Q2E: Query-to-Event Decomposition for Zero-Shot Multilingual
Collection
Datasets used in the paper: Q2E: Query-to-Event Decomposition for Zero-Shot Multilingual Text-to-Video Retrieval • 37 items • Updated
query stringlengths 16 163 | prequel listlengths 0 20 | sequel listlengths 0 20 | during listlengths 0 16 | metadata dict | num_of_frames int64 16 16 | frame_captions listlengths 16 16 | video_id stringlengths 9 9 | frame2video_caption stringlengths 141 2.1k |
|---|---|---|---|---|---|---|---|---|
a person is connecting something to system | [
"A person holding a cable or connector during connection to a system",
"A system or device with available ports or sockets for connection to a system",
"A person approaching the system with the cable or connector for connection to a system",
"A person inserting the cable or connector into the system during co... | [
"a person plugging in a cable to a computer port for connection to a system",
"A technician connecting wires to a circuit board during connection to a system",
"Someone inserting a USB drive into a laptop to establish a connection to a system",
"An engineer linking a device to a network router during connecti... | [
"A person holding a cable or connector for connection to a system",
"The person plugging the cable into a port or socket for connection to a system",
"the person turning on the device or system during connection to a system",
"Person configuring settings on the device or system during connection to a system",... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a person is connecting something to system<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: To extract events from the query \"a person is connecting something to system\", it makes sense to consider the physical actions involved in establishing a connection, as these can be visualized in a video. The events should focus on the specific steps or actions a person would take to connect an object or device to a system.\n\nEVENTS:\n1. A person holding a cable or connector\n2. The person plugging the cable into a port or socket\n3. The person turning on the device or system\n4. The person configuring settings on the device or system\n5. The person testing the connection to ensure it is working properly",
"en_article_excerpt": "a person is connecting something to system",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a person is connecting something to system<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are concrete and visual, such as physical actions or changes in the environment, that could lead to a person connecting something to a system. These events should be specific and relevant to the query, allowing for visualization in a video.\n\nEVENTS:\n1. A person holding a cable or connector\n2. A system or device with available ports or sockets\n3. A person approaching the system with the cable or connector\n4. A person inserting the cable or connector into the system\n5. The system's screen or indicators showing connection status or setup prompts",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a person is connecting something to system<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the action of connecting something to a system, considering the physical and visual aspects of the process. The events should be specific and plausible to be visualized in a video, taking into account common scenarios where connections to systems are made.\n\nEVENTS:\n1. A person plugging in a cable to a computer port\n2. A technician connecting wires to a circuit board\n3. Someone inserting a USB drive into a laptop\n4. An engineer linking a device to a network router\n5. A worker attaching a sensor to a machinery control system",
"video_URL": "https://youtube.com/watch?v=video9770",
"video_description": null,
"video_id": "video9770"
} | 16 | [
"A person is using a vacuum cleaner on a large tower computer case that is partially open. The case has the word \"hp\" on the side.",
"In the current frame, the person continues to use the vacuum cleaner on the large tower computer case. The case remains partially open, with the \"hp\" logo still visible.",
"I... | video9770 | A person is performing maintenance or repair on an HP tower computer. The process involves vacuuming the case and then manually adjusting and handling internal components. The person uses tools to secure or adjust parts inside the case, indicating a detailed and potentially complex repair or upgrade process. Throughout the activity, the computer case remains open, and the person's focus is on the internal components, suggesting a thorough and hands-on approach to the task. The video depicts a single, uninterrupted sequence of the person working on the computer, with no changes in setting or participants. |
a little girl does gymnastics | [
"little girl mounts a balance beam during a gymnastics performance",
"She performs a cartwheel on a mat during a gymnastics performance",
"girl hangs from uneven bars during a gymnastics performance",
"She executes a flip on a springboard during a gymnastics performance",
"The little girl strikes a pose aft... | [
"Cartwheeling across a mat during a gymnastics performance",
"Balancing techniques on a balance beam during a gymnastics performance",
"Performing a backflip on a springboard during a gymnastics performance",
"Twirling on the floor with a ribbon during a gymnastics performance",
"Landing a dismount from a m... | [
"A little girl running towards a springboard during a gymnastics performance",
"girl jumping over mini hurdle during gymnastics performance",
"girl performing a cartwheel on a mat during a gymnastics performance",
"girl balancing on a balance beam during a gymnastics performance",
"girl landing a dismount f... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a little girl does gymnastics<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: To decompose the query \"a little girl does gymnastics\" into simple events, it makes sense to extract concrete actions that are visually representative of gymnastics and can be reasonably captioned in a video frame. This will involve identifying common gymnastics movements that a little girl might perform, considering typical environments and equipment used in the sport.\n\nEVENTS:\n1. A little girl running towards a springboard\n2. The girl jumping over a mini hurdle\n3. The girl performing a cartwheel on a mat\n4. The girl balancing on a balance beam\n5. The girl landing a dismount from a mini trampoline",
"en_article_excerpt": "a little girl does gymnastics",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a little girl does gymnastics<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: To extract events related to the query \"a little girl does gymnastics\", it makes sense to consider common gymnastics movements and routines that a young girl might perform, as these would be visually captivating and concrete enough to be depicted in a video. The extracted events should be specific and relevant to the context of a child participating in gymnastics.\n\nEVENTS:\n1. The little girl mounts a balance beam\n2. She performs a cartwheel on a mat\n3. The girl hangs from uneven bars\n4. She executes a flip on a springboard\n5. The little girl strikes a pose after a routine",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a little girl does gymnastics<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are specific and concrete actions related to gymnastics that a little girl could be performing, as these would be easily visualized in a video. The events should be common gymnastics movements or routines that are suitable for a young girl.\n\nEVENTS:\n1. Cartwheeling across a mat\n2. Balancing on a balance beam\n3. Performing a backflip on a springboard\n4. Twirling on the floor with a ribbon\n5. Landing a dismount from a mini trampoline",
"video_URL": "https://youtube.com/watch?v=video9771",
"video_description": null,
"video_id": "video9771"
} | 16 | [
"A young gymnast is performing a routine on a gymnastics mat. The gymnast is in a dynamic pose, with one arm extended upwards and the other hand touching her hair, showcasing flexibility and balance. She is wearing a sleeveless leotard. The background shows a gym setting with a wall featuring horizontal stripes.",
... | video9771 | A young gymnast performs a dynamic routine on a blue gymnastics mat in a gym setting with horizontal stripes on the wall. The routine involves various movements, including poses, jumps, leaps, flips, and twists, showcasing her flexibility, balance, and control. Throughout the performance, the gymnast maintains focus and determination, demonstrating precision and skill in her technique. The routine includes moments of dynamic motion, such as running and jumping, as well as poised stances, highlighting her agility and readiness for each subsequent movement. The consistent gym setting provides a backdrop for her performance, which appears to be a practice session or a competition routine. The gymnast's confidence and control are evident throughout, as she transitions seamlessly between different parts of her routine, ultimately concluding in a relaxed and poised stance. |
a woman creating a fondant baby and flower | [
"Kneading fondant to soften it for creating a fondant baby",
"Kneading fondant to soften it for creating a flower",
"Shaping fondant into a baby figure for creating a fondant baby",
"Shaping fondant into a baby figure",
"Coloring fondant for flower petals when creating a fondant baby",
"Coloring fondant f... | [
"Kneading the fondant to soften it when creating a fondant baby",
"Kneading the fondant to soften it for creating a flower",
"Shaping the fondant into a baby figure while creating a fondant baby",
"Shaping the fondant into a baby figure",
"Creating intricate details on the baby's face when creating a fondan... | [
"A woman kneading fondant on a work surface while creating a fondant baby",
"A woman kneading fondant on a work surface while creating a flower",
"The woman shaping fondant into a baby figure while creating a fondant baby",
"The woman shaping fondant into a baby figure while creating a flower",
"The woman c... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a woman creating a fondant baby and flower<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: To extract events from the query \"a woman creating a fondant baby and flower\", it makes sense to break down the process of creating fondant decorations into individual, visualizable steps, considering the typical actions involved in working with fondant. This will involve identifying specific actions that can be reasonably captured in a video frame, such as molding, shaping, and decorating.\n\nEVENTS:\n1. A woman kneading fondant on a work surface\n2. The woman shaping fondant into a baby figure\n3. The woman creating petals for the flower out of fondant\n4. The woman assembling and decorating the fondant flower\n5. The woman adding final details to the fondant baby",
"en_article_excerpt": "a woman creating a fondant baby and flower",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a woman creating a fondant baby and flower<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: To extract events related to the query \"a woman creating a fondant baby and flower\", it makes sense to focus on the steps involved in the process of creating fondant decorations, as these would be concrete and visual events that could be captured in a video. The events should be specific to the task of shaping and decorating fondant to create a baby and flower, using common techniques and tools associated with fondant decoration.\n\nEVENTS:\n1. Kneading fondant to soften it\n2. Shaping fondant into a baby figure\n3. Coloring fondant for the flower petals\n4. Creating intricate details on the fondant baby's face\n5. Assembling the fondant flower petals into a complete flower",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a woman creating a fondant baby and flower<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are specific steps or moments in the process of creating a fondant baby and flower, as these would be concrete and visual elements that could be captured in a video. The events should be related to the creative process, such as shaping, molding, or decorating, and should be relevant to the subject matter of working with fondant.\n\nEVENTS:\n1. Kneading the fondant to soften it\n2. Shaping the fondant into a baby figure\n3. Creating intricate details on the baby's face\n4. Molding fondant into flower petals\n5. Assembling the flower and attaching it to the baby figurine",
"video_URL": "https://youtube.com/watch?v=video7020",
"video_description": null,
"video_id": "video7020"
} | 16 | [
"A person is making a craft or decoration using green and pink materials.",
"The person is shaping a green material around a pink object, which appears to be a small figurine, as part of their craft project.",
"The person continues to work on their craft project, carefully molding the green material around the ... | video7020 | A person is engaged in a crafting activity, initially working with green and pink materials to decorate a small figurine. They carefully shape and mold the green material around the pink figurine, adding leaf-shaped details to create a smooth and even appearance. The person then shifts their attention to working with dough, using a rolling pin to flatten a rectangular object on a grid-patterned surface. They proceed to cut out a flower shape from yellow dough using a white flower-shaped cutter and then trim the edges of the cut-out flower using purple scissors, refining its shape. Throughout the video, the person is focused on precision and detail, working diligently to create their craft projects. |
a boy plays grand theft auto 5 | [
"A boy sitting in front of a TV or computer with a controller in his hand playing a video game",
"The boy playing a video game and navigating the menu to start a new game or load a saved file",
"The boy's character in the game walking or driving through the virtual city of Los Santos while playing a video game"... | [
"The boy holding a controller and focusing on the screen while playing a video game",
"The boy's character in a video game driving a car and crashing into other vehicles",
"The boy's character engaging in a shootout with non-player characters while playing a video game",
"The boy playing a video game and navi... | [
"A boy sitting in front of a TV or computer screen playing a video game",
"The boy holding a gaming controller while playing a video game",
"Scenes from Grand Theft Auto 5 displayed on the screen while playing the video game, such as driving cars or interacting with characters",
"The boy's hands moving the co... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a boy plays grand theft auto 5<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: To extract events from the query \"a boy plays Grand Theft Auto 5\", it makes sense to focus on the actions and scenes that are likely to occur while someone is playing the game, as well as the environment in which the game is being played. This will involve considering common aspects of the game and the setting in which a boy would typically play it.\n\nEVENTS:\n1. A boy sitting in front of a TV or computer screen\n2. The boy holding a gaming controller\n3. Scenes from Grand Theft Auto 5 displayed on the screen, such as driving cars or interacting with characters\n4. The boy's hands moving the controller to control gameplay\n5. The boy reacting to the game, such as looking excited or concentrated",
"en_article_excerpt": "a boy plays grand theft auto 5",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a boy plays grand theft auto 5<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the act of playing a video game, specifically Grand Theft Auto 5, as these events would likely be visualized in a video of someone playing the game. The extracted events should focus on the boy's interactions with the game and the game's environment.\n\nEVENTS:\n1. A boy sitting in front of a TV or computer with a controller in his hand\n2. The boy navigating the game's menu to start a new game or load a saved file\n3. The boy's character in the game walking or driving through the virtual city of Los Santos\n4. The boy engaging in a shootout or car chase within the game\n5. The boy pausing the game or adjusting the settings on his console or computer",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a boy plays grand theft auto 5<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: Given the query \"a boy plays Grand Theft Auto 5\", it makes sense to extract events that are directly related to the gameplay or the environment in which the boy is playing, as these would be the most visual and relevant aspects of a video on this topic. The events should reflect common scenarios or actions that occur while playing the game or interacting with the gaming setup.\n\nEVENTS:\n1. The boy holding a controller and focusing on the screen\n2. The boy's character in the game driving a car and crashing into other vehicles\n3. The boy's character engaging in a shootout with non-player characters\n4. The boy navigating the game's menu to select a mission or customize his character\n5. The boy reacting emotionally to a intense or surprising moment in the game",
"video_URL": "https://youtube.com/watch?v=video9773",
"video_description": null,
"video_id": "video9773"
} | 16 | [
"Several motorcyclists are riding on a wet street at night, with their headlights on. They appear to be moving at a moderate speed.",
"The current frame displays the date \"MARCH 10\" against a black background, likely indicating a transition or a title card within the video. There are no visible motorcyclists or... | video9773 | The video depicts a racing scenario, primarily focusing on a green sports car navigating through a curvy, rural road in a hilly landscape. The car participates in a high-speed race, complete with checkpoints and ramps for performing aerial stunts. At one point, the car launches off a ramp, soaring over the hilly terrain and even a cityscape, reaching high speeds during its airborne sequence. The car eventually lands back on the road, continuing its high-speed journey. The video also includes a brief scene of motorcyclists riding at night and a title card displaying the date "MARCH 10," which may set the context or serve as a narrative pause. The majority of the video, however, centers around the green sports car's racing and stunt performance, captured in what appears to be a gameplay scenario from a racing video game. |
a man is giving a review on a vehicle | [
"The man walking around the vehicle to point out its features during a review of a vehicle",
"The man sitting in the driver's seat and adjusting the controls during a review of a vehicle",
"The man starting the engine and listening to its sound during a review of a vehicle",
"The man driving the vehicle to te... | [
"The man walking around the vehicle to inspect it during a review of a vehicle",
"The man sitting in the driver's seat and adjusting the controls during a review of a vehicle",
"Review of a vehicle's trunk space and features",
"The man test-driving the vehicle on the road during a review of a vehicle",
"The... | [
"A man standing next to a vehicle, pointing out its features during a review of a vehicle",
"The man sitting in the driver's seat, testing the vehicle's controls during a review of a vehicle",
"The man walking around the vehicle, inspecting its exterior during a review of a vehicle",
"The man driving the vehi... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man is giving a review on a vehicle<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the act of reviewing a vehicle, such as the man interacting with the vehicle or expressing his opinions. These events should be concrete and visual, allowing them to be easily depicted in a video.\n\nEVENTS:\n1. A man standing next to a vehicle, pointing out its features\n2. The man sitting in the driver's seat, testing the vehicle's controls\n3. The man walking around the vehicle, inspecting its exterior\n4. The man driving the vehicle, testing its performance\n5. The man speaking directly to the camera, sharing his thoughts on the vehicle",
"en_article_excerpt": "a man is giving a review on a vehicle",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man is giving a review on a vehicle<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: To extract events related to the query \"a man is giving a review on a vehicle\", it makes sense to consider the typical actions and scenes that would be involved in a video review of a vehicle, such as the reviewer interacting with the vehicle and sharing their opinions. The extracted events should be concrete and visual, allowing them to be easily captioned and depicted in a video frame.\n\nEVENTS:\n1. The man walking around the vehicle to point out its features\n2. The man sitting in the driver's seat and adjusting the controls\n3. The man starting the engine and listening to its sound\n4. The man driving the vehicle to test its performance\n5. The man standing in front of the vehicle, speaking to the camera about its pros and cons",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man is giving a review on a vehicle<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the act of reviewing a vehicle, such as the man interacting with the vehicle or expressing his opinion. These events should be concrete and visual, allowing them to be easily captioned and depicted in a video.\n\nEVENTS:\n1. The man walking around the vehicle to inspect it\n2. The man sitting in the driver's seat and adjusting the controls\n3. The man checking the vehicle's trunk space and features\n4. The man test-driving the vehicle on the road\n5. The man speaking directly to the camera to share his thoughts on the vehicle",
"video_URL": "https://youtube.com/watch?v=video7026",
"video_description": null,
"video_id": "video7026"
} | 16 | [
"The current frame depicts a vehicle driving through a rocky, natural landscape with large boulders and sparse vegetation. The car's hood is visible in the foreground, suggesting that the perspective is from inside the vehicle.",
"The vehicle continues to drive through the rocky, natural landscape, maintaining it... | video7026 | A vehicle, driven by a man, navigates through a rocky, natural landscape with large boulders and sparse vegetation. The driver prepares for a controlled descent, as indicated by the dashboard displays, and the vehicle momentarily remains stationary in park mode. During this time, the driver engages in conversation, gesturing with his hands while explaining or demonstrating something related to the vehicle. After the conversation, the driver proceeds to drive the vehicle, focusing on the road as the speedometer shows the car is moving. The video depicts the driver's actions and the vehicle's movement through the natural terrain, highlighting the preparation and execution of a controlled descent. |
a man speaks to children in a classroom | [
"A man stands at the front of the classroom writing on a blackboard while speaking to the children",
"Children sit at their desks looking up at the man as he speaks to them",
"A man speaking to children while gesturing",
"A man speaking to children as they raise their hands to ask questions",
"The man walks... | [
"A man writes on a blackboard while speaking to children",
"Children take notes at their desks while a man speaks to them",
"A man gestures as he explains a concept to a group of children while speaking to them",
"Students ask questions and engage in discussion when a man speaks to them",
"A man speaking to... | [
"A man stands at the front of the classroom writing on the blackboard while speaking to the children",
"Children sit at their desks listening to the man speak",
"A man speaking to children as he walks around the classroom to assist individual students",
"Students raise their hands to ask the man questions dur... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man speaks to children in a classroom<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that are typical of a classroom setting and involve interaction between a teacher and students, as these are common and visual aspects of such a scene. The events should be specific and concrete enough to be captured in a video, focusing on the actions and activities that occur in a classroom.\n\nEVENTS:\n1. A man stands at the front of the classroom writing on the blackboard\n2. Children sit at their desks listening to the man speak\n3. The man walks around the classroom to assist individual students\n4. Students raise their hands to ask the man questions\n5. The man uses visual aids such as a projector or diagrams to explain a concept",
"en_article_excerpt": "a man speaks to children in a classroom",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man speaks to children in a classroom<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are typical of a classroom setting and involve interaction between a teacher and students, as these are common and visual elements that could be captured in a video. The events should be specific and relevant to the context of a man speaking to children in a classroom, avoiding generic or unrelated events.\n\nEVENTS:\n1. A man stands at the front of the classroom writing on a blackboard\n2. Children sit at their desks looking up at the man\n3. The man gestures as he speaks to the children\n4. Children raise their hands to ask questions\n5. The man walks around the classroom to assist individual students",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man speaks to children in a classroom<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are typical of a classroom setting and involve interaction between a teacher and students, as the query suggests a educational or instructional context. The events should be concrete and visual, allowing for a clear mental image of what the scene might look like in a video.\n\nEVENTS:\n1. A man writes on a blackboard\n2. Children take notes at their desks\n3. The man gestures as he explains a concept\n4. Students ask questions and engage in discussion\n5. The man hands out worksheets or assignments to the children",
"video_URL": "https://youtube.com/watch?v=video9775",
"video_description": null,
"video_id": "video9775"
} | 16 | [
"The current frame shows a group of children in a classroom setting, likely engaged in a learning or storytelling activity. They are seated and appear to be listening attentively, with some showing expressions of interest. The children are wearing name tags, indicating a structured educational or camp environment. ... | video9775 | A group of children, wearing name tags, are engaged in a learning or storytelling activity in a classroom setting. They appear to be listening attentively and participating in the discussion. A man, likely a teacher or educational administrator, is present in the classroom, observing and taking notes. He engages in conversation with the children, providing instructions or guidance. The children respond positively, showing interest and attentiveness. The classroom environment is filled with educational materials, posters, and colorful storage bins, reinforcing the learning setting. The video depicts an interactive and engaging educational activity, where the children are encouraged to participate and respond to the discussion, under the guidance of the man. |
one micky mouse is talking to other | [
"Mickey Mouse standing in front of a group of characters during a conversation",
"Mickey Mouse shaking hands with another character during a conversation",
"Mickey Mouse having a conversation with Minnie Mouse",
"Mickey Mouse gesturing to a group of Disney characters during a conversation",
"Mickey Mouse po... | [
"Mickey Mouse waving at other Disney characters during a conversation",
"Mickey Mouse having a conversation and shaking hands with other Disney characters",
"Mickey Mouse having a conversation with Minnie Mouse during their conversation scene",
"Mickey Mouse laughing with Goofy and Pluto during a conversation... | [
"Mickey Mouse standing in front of other characters during a conversation",
"Mickey Mouse waving at other characters during a conversation",
"Other characters gathering around Mickey Mouse during a conversation",
"Mickey Mouse pointing to something and talking during a conversation",
"other characters laugh... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: one micky mouse is talking to other<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that are typical of a scene where Mickey Mouse is interacting with other characters, likely in a cartoon or animated setting. The events should be simple, visual, and concrete, such as gestures, movements, or actions that can be easily depicted in a video frame.\n\nEVENTS:\n1. Mickey Mouse standing in front of other characters\n2. Mickey Mouse waving at other characters\n3. Other characters gathering around Mickey Mouse\n4. Mickey Mouse pointing to something and talking\n5. Other characters laughing and responding to Mickey Mouse",
"en_article_excerpt": "one micky mouse is talking to other",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: one micky mouse is talking to other<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: Given the query \"one Mickey Mouse is talking to other\", it makes sense to extract events that are related to Mickey Mouse interacting with other characters, likely in a cartoon or animated setting. The events should be concrete and visual, allowing for easy depiction in a video.\n\nEVENTS:\n1. Mickey Mouse standing in front of a group of characters\n2. Mickey Mouse shaking hands with another character\n3. Mickey Mouse having a conversation with Minnie Mouse\n4. Mickey Mouse gesturing to a group of Disney characters\n5. Mickey Mouse pointing to a map or object while explaining to others",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: one micky mouse is talking to other<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: Given the query \"one Mickey Mouse is talking to other\", it makes sense to extract events that are typical of a scene where Mickey Mouse is interacting with other characters, likely in a cartoon or theme park setting. The events should be concrete and visual, such as physical interactions or reactions between Mickey Mouse and the other characters.\n\nEVENTS:\n1. Mickey Mouse waving at other Disney characters\n2. Mickey Mouse shaking hands with other characters\n3. Mickey Mouse having a conversation with Minnie Mouse\n4. Mickey Mouse laughing with Goofy and Pluto\n5. Mickey Mouse giving a high-five to Donald Duck",
"video_URL": "https://youtube.com/watch?v=video9776",
"video_description": null,
"video_id": "video9776"
} | 16 | [
"The image shows a cartoon character that resembles a cow with a happy expression, wearing a large yellow hat and a yellow bow tie. The background features a pink and purple color scheme with some abstract shapes and a part of a clock visible on the left side.",
"The cartoon cow character, still donning its yello... | video9776 | A cartoon cow character, wearing a yellow hat and bow tie, is the main participant in the video. The character maintains a cheerful expression throughout, engaging in what appears to be a lively conversation or song. The setting features a consistent pink and purple color scheme with abstract shapes and a partial view of a clock in the background. As the video progresses, the cow character introduces a wand with a star on top, using it to perform a magic trick or gesture, which adds a magical theme to the scene. The character's expressions vary from cheerful to surprised or curious, but the overall atmosphere remains cheerful and animated. The video depicts the cow character's enthusiastic and magical performance, with no other participants or significant changes in the setting. |
a naked child runs through a field | [
"A child taking off their clothes to run through a field and play in a sprinkler or water hose on a hot day.",
"A family having a picnic in a field with a child running freely after removing their clothes",
"A child playing a game of tag or hide-and-seek while running through a field and shedding clothes.",
"... | [
"A child running through a field of tall grass",
"A child laughs and spins around while running through a field",
"A child running through a field chasing after a butterfly or insect",
"child falls and gets up while running through a field",
"A child running through a field reaches the edge and climbs over ... | [
"A child runs through a field",
"A child running through a field while laughing and spinning around",
"A child running through a field, picking wildflowers or touching the grass",
"A child running through a field trips and falls, then gets back up",
"A child running through a field, smiling and unselfconsci... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a naked child runs through a field<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: This query describes a simple and vivid scene that can be easily visualized in a video, and it makes sense to extract events that focus on the child's actions and the setting. The events should capture the carefree and natural atmosphere of a child playing in a field, and can be reasonably captioned as individual frames in a video.\n\nEVENTS:\n1. A child runs through a field\n2. The child laughs and spins around in the field\n3. The child picks wildflowers or touches the grass in the field\n4. The child trips and falls in the field, then gets back up\n5. The child runs towards the camera, smiling and unselfconscious",
"en_article_excerpt": "a naked child runs through a field",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a naked child runs through a field<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: Given the query \"a naked child runs through a field\", it makes sense to extract events that could lead to a child being in a state of undress and in a natural setting, considering typical scenarios where a child might be without clothes and in a field. These events should be concrete and visualizable in a video, taking into account common activities or situations involving children in outdoor settings.\n\nEVENTS:\n1. A child taking off their clothes to play in a sprinkler or water hose on a hot day.\n2. A family having a picnic in a field, with the child removing their clothes to run around freely.\n3. A child playing a game of tag or hide-and-seek in a field, shedding clothes in the process of playing.\n4. A child swimming in a nearby pond or lake and then running through the field to get back to their towel or clothes.\n5. A family participating in a music festival or a nudist event in a field, where the child is naturally without clothes.",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a naked child runs through a field<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: Given the query \"a naked child runs through a field\", it makes sense to extract events that are related to the child's actions and interactions with the field environment, considering typical scenarios that could be visually depicted in a video. The events should be concrete and plausible, taking into account common activities and settings associated with a child running in a field.\n\nEVENTS:\n1. The child runs through tall grass\n2. The child laughs and spins around in the field\n3. The child chases after a butterfly or insect\n4. The child falls and gets up, continuing to run\n5. The child reaches the edge of the field and climbs over a fence",
"video_URL": "https://youtube.com/watch?v=video7025",
"video_description": null,
"video_id": "video7025"
} | 16 | [
"A young child is running on a grassy field near a body of water, with other people visible in the background.",
"A woman is running on a grassy field near a body of water, with other people visible in the background.",
"A woman is walking on a grassy field near a body of water, with people lounging and enjoyin... | video7025 | The video depicts two distinct scenes. The first part shows a leisurely day at a beach near a body of water, where people are lounging, sunbathing, and enjoying the sunny day. The scene transitions through various individuals, including a young child and a woman, before focusing on a group of people relaxing on the beach. The atmosphere is serene, with trees and foliage providing a natural backdrop.
The second part of the video abruptly shifts to a series of black and white photographs of a young man in a uniform, suggesting a formal or professional setting, possibly military or official. The photographs are formal portraits, with the individual maintaining a serious expression and looking directly at the camera. The focus remains solely on the individual, with no additional context or background details visible. The video ends with a close-up of the young man's face, still in the formal portrait context. |
a little boy singing in front of judges and crowd | [
"A little boy standing on a stage with a microphone during a singing performance",
"A little boy standing on a stage with a microphone during a talent show or audition",
"Judges seated at a table with scorecards and pens during a singing performance",
"Judges seated at a table with scorecards and pens at a ta... | [
"The little boy stepping up to the microphone during a singing performance",
"The little boy stepping up to the microphone at a talent show or audition",
"The crowd applauding as the little boy begins to sing at his singing performance",
"The crowd applauding as the little boy begins to sing at a talent show ... | [
"A little boy standing on a stage with a microphone during a singing performance",
"A little boy standing on a stage with a microphone during a talent show or audition",
"boy giving an emotional singing performance",
"boy singing with emotional expression on a talent show or audition",
"Judges looking at th... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a little boy singing in front of judges and crowd<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: Given the query \"a little boy singing in front of judges and crowd\", it makes sense to extract events that are typical of a singing competition or audition setting, where a young performer is showcasing their talent to a panel of judges and an audience. The extracted events should be concrete and visual, allowing for easy depiction in a video.\n\nEVENTS:\n1. A little boy standing on a stage with a microphone\n2. The boy singing with emotional expression\n3. Judges looking at the boy with focused attention\n4. The crowd applauding and cheering\n5. The boy taking a bow after finishing his song",
"en_article_excerpt": "a little boy singing in front of judges and crowd",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a little boy singing in front of judges and crowd<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are typical of a talent show or competition setting, as the query suggests a performance in front of an audience and judges. The events should be concrete and visual, allowing for a clear mental image of what the scene might look like in a video.\n\nEVENTS:\n1. A little boy standing on a stage with a microphone\n2. Judges seated at a table with scorecards and pens\n3. A crowd of people cheering and applauding in the audience\n4. The little boy taking a deep breath before starting to sing\n5. Confetti and balloons falling from the ceiling after the performance",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a little boy singing in front of judges and crowd<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: Given the query \"a little boy singing in front of judges and crowd\", it makes sense to extract events that are typical of a singing competition or audition setting, as the presence of judges and a crowd suggests a performance-based scenario. The extracted events should be concrete and visual, focusing on the actions of the little boy and the reactions of the judges and crowd.\n\nEVENTS:\n1. The little boy stepping up to the microphone\n2. The crowd applauding as the little boy begins to sing\n3. Judges taking notes and looking impressed\n4. The little boy performing an emotional or powerful part of the song\n5. The crowd giving a standing ovation after the little boy finishes singing",
"video_URL": "https://youtube.com/watch?v=video9778",
"video_description": null,
"video_id": "video9778"
} | 16 | [
"A woman with curly hair is sitting on a red chair, smiling and looking down, possibly at notes or a script on a desk in front of her.",
"The woman with curly hair is now standing, appearing to adjust her jacket or blouse, possibly preparing to speak or move. The audience in the background remains attentive.",
... | video9778 | A woman with curly hair is initially preparing to speak on a stage, adjusting her attire and engaging with the audience. She then transitions, allowing a young boy with curly hair to take the stage, where he addresses the audience using a microphone. The boy's segment appears to conclude with him clapping, acknowledging the audience. The scene then shifts to a man in a red chair, wearing a "GUILTY" t-shirt and sunglasses, who is clapping enthusiastically. The video culminates with three individuals on stage in front of "The Voice" logo, where one is singing, and the others are clapping and showing appreciation, suggesting a musical performance or competition setting. |
fireworks are being lit and exploding in a night sky | [
"People setting up fireworks on the ground at night that lead to a fireworks explosion",
"People setting up fireworks on the ground at night for a fireworks display",
"A person lighting a firework fuse at night leading to a fireworks explosion",
"A person lighting a firework fuse during a nighttime fireworks ... | [
"Fireworks shooting up into the air during a fireworks explosion at night",
"Fireworks shooting up into the air during a nighttime fireworks display",
"Explosions of colorful lights in the night sky from fireworks explosion",
"Fireworks display explosions of colorful lights in the night sky",
"sparks and em... | [
"Fireworks explosion at night with fireworks shooting up into the night sky",
"Fireworks display at night with fireworks shooting up into the night sky",
"Explosions of colorful lights and patterns in the sky at night from fireworks displays",
"Explosions of colorful lights and patterns in the sky during a ni... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: fireworks are being lit and exploding in a night sky<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that describe the visual and auditory aspects of fireworks being lit and exploding in a night sky, as these events can be easily visualized in a video. The extracted events should focus on the fireworks themselves, as well as the surrounding environment and potential reactions from people watching.\n\nEVENTS:\n1. Fireworks shooting up into the night sky\n2. Explosions of colorful lights and patterns in the sky\n3. Sparks and embers falling back down to the ground\n4. People watching and reacting to the fireworks display\n5. Silhouettes of buildings or trees against the bright night sky",
"en_article_excerpt": "fireworks are being lit and exploding in a night sky",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: fireworks are being lit and exploding in a night sky<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the setup and execution of fireworks, as these events are likely to be visualized in a video associated with the query. Events such as preparation, ignition, and explosion of fireworks are concrete and can be reasonably captioned in a video frame.\n\nEVENTS:\n1. People setting up fireworks on the ground\n2. A person lighting a firework fuse\n3. Fireworks shooting up into the night sky\n4. Fireworks exploding into colorful patterns\n5. Sparks and debris falling back down to the ground",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: fireworks are being lit and exploding in a night sky<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are visually related to fireworks and their explosion in a night sky, considering the typical setting and characteristics of such an event. The extracted events should be concrete and likely to be seen in a video capturing fireworks explosions.\n\nEVENTS:\n1. Fireworks shooting up into the air\n2. Explosions of colorful lights in the night sky\n3. Sparks and embers falling back down to the ground\n4. People watching and reacting to the fireworks display\n5. Fireworks exploding in synchronized patterns and designs",
"video_URL": "https://youtube.com/watch?v=video9779",
"video_description": null,
"video_id": "video9779"
} | 16 | [
"Fireworks exploding in the night sky, with colorful flags visible on a ship's mast.",
"Fireworks continue to light up the night sky, with a prominent burst of golden sparks near the top of the frame. A ship's mast with colorful flags remains visible, adding to the festive atmosphere.",
"Fireworks illuminate th... | video9779 | A fireworks display is taking place at night, near a ship with a mast adorned with colorful flags. The fireworks explode in the sky, creating a festive and celebratory atmosphere, with bursts of golden sparks and vibrant colors illuminating the dark surroundings. The ship's mast with colorful flags remains visible throughout the display, contributing to the overall festive mood. The fireworks continue to explode and expand in the sky, forming intricate patterns with sparkling lights, and eventually reaching a peak with a large, bright, and colorful burst. The display maintains its vibrancy and celebratory atmosphere until the end, with the ship's presence and the fireworks combining to create a joyful and festive event. |
a man is singing and standing in the road | [
"A man picks up a guitar case and walks out of a building to go singing",
"A man picks up a guitar case and walks out of a building while standing",
"A man walks down the street looking for a spot to perform his singing",
"A man walks down the street, looking for a spot to stand and perform",
"A man stands ... | [
"The man sings to a crowd of people gathered around him",
"The man stands and sings to a crowd of people gathered around him",
"Cars drive by the man while he is singing in the road",
"Cars drive by the man while he is standing and singing in the road",
"people stopping to listen to a man singing in the str... | [
"A man stands in the middle of the road singing",
"A man stands in the middle of the road",
"The man sings with his mouth open",
"The man sings with his mouth open while standing",
"people passing by look at the man singing",
"people passing by look at the man standing and singing",
"Cars drive around t... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man is singing and standing in the road<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the actions of the man, such as his singing and standing in the road, as these are concrete and visual events that could be captured in a video. The events should focus on the specific actions and setting described in the query, without adding extraneous details.\n\nEVENTS:\n1. A man stands in the middle of the road\n2. The man sings with his mouth open\n3. People passing by look at the man singing\n4. Cars drive around the man standing in the road\n5. The man gestures with his hands while singing",
"en_article_excerpt": "a man is singing and standing in the road",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man is singing and standing in the road<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: Given the query \"a man is singing and standing in the road\", it makes sense to extract events that could lead up to this scene, such as the man's preparation or actions before standing in the road to sing. These events should be concrete and visual, allowing them to be easily depicted in a video.\n\nEVENTS:\n1. A man picks up a guitar case and walks out of a building\n2. A man walks down the street, looking for a spot to perform\n3. A man stands at the side of the road, takes a deep breath, and begins to sing\n4. A man sets up a microphone stand in the middle of the road\n5. A man attracts a crowd of people who stop to listen to him sing",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: a man is singing and standing in the road<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the actions of the man singing and standing in the road, considering the context of a video that could visually capture these moments. The events should be concrete and plausible, such as interactions with passersby, reactions from the audience, or the man's performance itself.\n\nEVENTS:\n1. The man sings to a crowd of people gathered around him\n2. Cars drive by the man while he is singing in the road\n3. People stop to listen to the man singing\n4. The man plays a musical instrument while singing in the road\n5. Bystanders throw money into a tip jar or hat in front of the man",
"video_URL": "https://youtube.com/watch?v=video7028",
"video_description": null,
"video_id": "video7028"
} | 16 | [
"A person is skateboarding in front of a traditional Japanese gate, with several people standing and watching in the background.",
"A person is walking past a traditional Japanese gate, wearing sunglasses and a beanie. The background shows a crowd of people and a large red lantern hanging in front of the gate.",
... | video7028 | The video depicts a series of scenes showcasing a person's experiences in a Japanese cultural setting, interspersed with urban landscapes. The video begins with the person skateboarding in front of a traditional Japanese gate, followed by scenes of them walking and interacting with others near the gate. They are then shown participating in a lively celebration at a traditional Japanese bar or restaurant, toasting with a group of friends. The group continues their social interaction in an open area with traditional architectural elements, before the scene shifts to an urban setting where the person is seen looking at a sign and later attending a festive event with others. The video also features scenes of the person walking in busy city streets at night, interacting with the camera, and visiting a traditional temple. Throughout the video, the person is often shown wearing casual, eclectic clothing, including a beanie, sunglasses, and a necklace. The overall context suggests a cultural exchange or travel experience, with the person exploring and engaging with Japanese culture and urban environments. |
cartoon show for kids | [
"Colorful characters singing and dancing together in a cartoon show",
"a group of friends playing a game or sport in a fantastical environment in a cartoon show",
"A young hero character learning a valuable lesson from a wise mentor in a cartoon show",
"A comedic villain getting outsmarted by the main charact... | [
"colorful characters dancing and singing together in a cartoon show",
"A group of animal friends playing in a magical forest in a cartoon show",
"A superhero cartoon character saving the day in a cartoon show",
"A cartoon teacher explaining a fun science experiment on a cartoon show",
"A group of kids on a ... | [
"Colorful characters singing and dancing together in a cartoon show",
"a group of cartoon friends playing a game or sport in a cartoon show",
"A cartoon character learning a new skill or lesson from a teacher in a cartoon show",
"Cartoon animals going on a fun adventure or exploration in a cartoon show",
"A... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: cartoon show for kids<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: Given the query \"cartoon show for kids\", it makes sense to extract events that are commonly found in children's cartoons, such as characters interacting with each other, engaging in fun activities, or learning valuable lessons. These events should be concrete and visually engaging, making them suitable for a video caption.\n\nEVENTS:\n1. Colorful characters singing and dancing together\n2. A group of cartoon friends playing a game or sport\n3. A cartoon character learning a new skill or lesson from a teacher\n4. Cartoon animals going on a fun adventure or exploration\n5. A birthday party or celebration with cartoon characters and kids",
"en_article_excerpt": "cartoon show for kids",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: cartoon show for kids<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: Given the query \"cartoon show for kids\", it makes sense to extract events that are commonly found in children's cartoons, such as characters interacting with each other, engaging in fun activities, or learning valuable lessons. These events should be colorful, lively, and easy to visualize in a video format.\n\nEVENTS:\n1. Colorful characters singing and dancing together\n2. A group of friends playing a game or sport in a fantastical environment\n3. A young hero character learning a valuable lesson from a wise mentor\n4. A comedic villain getting outsmarted by the main characters\n5. A fun and imaginative adventure sequence with characters exploring a new world",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: cartoon show for kids<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: Given the query \"cartoon show for kids\", it makes sense to extract events that are typical scenes or plot elements commonly found in children's cartoons, such as character interactions, action sequences, or educational moments. These events should be visually engaging and easy to imagine as a video clip.\n\nEVENTS:\n1. Colorful characters dancing and singing together\n2. A group of animal friends playing in a magical forest\n3. A superhero cartoon character saving the day\n4. A cartoon teacher explaining a fun science experiment\n5. A group of kids on a cartoon adventure, exploring a new planet",
"video_URL": "https://youtube.com/watch?v=video7029",
"video_description": null,
"video_id": "video7029"
} | 16 | [
"A cartoon character resembling a blue octopus with a pink hat and an orange shirt is standing next to a wall-mounted telephone. The character appears to be reaching out to answer the phone.",
"The blue octopus character is now holding the phone's receiver, seemingly engaged in a conversation, as indicated by its... | video7029 | A video depicts a conversation between a blue octopus character and a yellow sponge character over the telephone. The blue octopus answers the phone and engages in a conversation, displaying focused and attentive expressions. Meanwhile, the yellow sponge character initially reacts with surprise, then joins the conversation, smiling and appearing to have a pleasant exchange. However, the yellow sponge's mood shifts, and he becomes frustrated and distressed, displaying anger and pulling his hair. The situation resolves as the yellow sponge character returns to speaking on the phone, this time showing a significant improvement in mood, smiling, and looking relaxed throughout the remainder of the conversation. The context suggests a lighthearted and comedic exchange between the two characters, with the yellow sponge experiencing a range of emotions before ultimately returning to a positive state. |
some cartoon characters are moving around an area | [
"Cartoon characters walking in different directions",
"Characters running into each other and reacting during the movement of cartoon characters",
"Characters interacting with objects in the area, such as picking up items or climbing structures during the movement of cartoon characters",
"Cartoon characters e... | [
"cartoon characters running around in circles",
"Characters jumping over obstacles in the movement of cartoon characters",
"Characters interacting with each other through conversation or gestures in the movement of cartoon characters",
"cartoon characters chasing one another in animated movement sequences",
... | [
"cartoon characters walking around a park in motion or movement scenes",
"cartoon characters running through a city street",
"cartoon characters jumping over obstacles in a forest during movement sequences",
"Characters interacting with each other through movement in a town square",
"Cartoon characters danc... | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: some cartoon characters are moving around an area<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that are visually concrete and related to the movement of cartoon characters in a given area, considering typical scenarios often depicted in cartoons. This could involve various actions and interactions between characters, as well as movements within their environment.\n\nEVENTS:\n1. Cartoon characters walking around a park\n2. Characters running through a city street\n3. Characters jumping over obstacles in a forest\n4. Characters interacting with each other in a town square\n5. Characters dancing in a colorful meadow",
"en_article_excerpt": "some cartoon characters are moving around an area",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: some cartoon characters are moving around an area<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are typical of cartoon characters and their interactions with their environment, considering common cartoon scenarios and settings. The extracted events should be visually engaging and plausible in a cartoon context, such as characters walking, running, or interacting with objects.\n\nEVENTS:\n1. Cartoon characters walking in different directions\n2. Characters running into each other and reacting\n3. Characters interacting with objects in the area, such as picking up items or climbing structures\n4. Characters entering and exiting buildings or other enclosed spaces\n5. Characters using transportation like bicycles or skateboards to move around",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: some cartoon characters are moving around an area<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are typical actions or behaviors associated with cartoon characters in a dynamic setting, such as interacting with each other or their environment. Given the vague nature of the query, the extracted events should be general and common in cartoon scenarios.\n\nEVENTS:\n1. Cartoon characters running around in circles\n2. Characters jumping over obstacles\n3. Characters interacting with each other through conversation or gestures\n4. Characters chasing one another\n5. Characters playing with objects or toys in the area",
"video_URL": "https://youtube.com/watch?v=video9772",
"video_description": null,
"video_id": "video9772"
} | 16 | [
"A character with wings is flying above a rocky landscape.",
"A character with wings is continuing to fly over the rocky landscape, maintaining their forward momentum.",
"The character with wings continues to soar over the rocky terrain, maintaining their forward momentum, and appears to be getting closer to th... | video9772 | A winged character flies over a rocky landscape, navigating through a narrow canyon and eventually reaching a coastal area with a castle-like structure. The character lands on a rocky outcrop and interacts with two individuals on the beach, before taking flight again. As the character soars higher into the sky, the two individuals watch from the shore, set against a serene coastal backdrop. The character continues to ascend, eventually disappearing into the clouds. The video then shifts to depict a fantastical urban landscape, featuring a city with a prominent building, suggesting a new setting or destination, possibly related to the character's journey. |
baseball player hits ball | [
"Pitcher throwing the ball to hit it",
"batter swinging the bat while hitting a ball",
"Ball flying through the air after being hit during a hit",
"runner rounding the bases after hitting a ball",
"Fielder attempting to catch the hit ball during a baseball game after a batter is hitting a ball"
] | [
"Baseball player swinging bat while hitting a ball",
"ball flying through the air when hitting a ball",
"Ball landing in the outfield after hitting a ball",
"Runner rounding bases after hitting a ball",
"Fielder catching the ball after hitting it"
] | [
"Baseball player swings bat while hitting a ball",
"ball leaves pitcher's hand when hitting a ball",
"Bat makes contact with ball when hitting a ball",
"Ball soars through the air when hitting a ball",
"Player runs to first base after hitting a ball"
] | {
"article_url": null,
"during_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to decompose the ongoing event in multiple simple events. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: baseball player hits ball<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"during_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the action of hitting a ball in a baseball game, as these events are likely to be visualized in a video of the query. The extracted events should focus on the physical actions involved in hitting a ball, such as the player's swing and the ball's movement.\n\nEVENTS:\n1. Baseball player swings bat\n2. Ball leaves pitcher's hand\n3. Bat makes contact with ball\n4. Ball soars through the air\n5. Player runs to first base",
"en_article_excerpt": "baseball player hits ball",
"en_article_url": null,
"event_category": null,
"event_name": null,
"language": null,
"prequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what previous events could lead to this given event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: baseball player hits ball<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"prequel_raw_generation": "EXPLANATION: It makes sense to extract events that are directly related to the action of hitting a ball in a baseball game, as these events are likely to be visualized in a video of the query. The extracted events should be specific to the sport of baseball and the action of hitting, taking into account common scenarios that occur during a game.\n\nEVENTS:\n1. Pitcher throwing the ball\n2. Batter swinging the bat\n3. Ball flying through the air after being hit\n4. Runner rounding the bases\n5. Fielder attempting to catch the hit ball",
"sequel_prompt": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\nCutting Knowledge Date: December 2023\nToday Date: 26 Jul 2024\n\n<|eot_id|><|start_header_id|>user<|end_header_id|>\n\nYou are given a video search query. Assume the video search query is an ongoing event. You have to find out what events can be the outcome of this ongoing event. The extracted events should be concrete enough to be visualized in a video.\n\nHere are some guidelines: \n- Extract events that could be visualized in a video.\n- Before extracting an event, consider whether the event text could serve as a reasonable caption for a frame of a video associated with the query.\n- Give specific events related to the query, use your previous WORLD KNOWLEDGE to extract events that are likely to be seen in a video. For example, if the query is \"2019 Christmas Eve Protests in Hong Kong\", you might return \"Protesters marching in the streets\", but not \"Protestors marching in snow\" because it is not likely to have snow in Hong Kong, though its pretty common in other places. \n- First, explain your reasoning in 1-2 sentences about which events it makes sense to extract. Then, provide a numbered list with the events (anywhere from one to five event; use your best judgement to determine how many to extract)\n- Structure your response into two sections, \"\"EXPLANATION:\"\" and then \"\"EVENTS:\"\". The events sections should contain nothing but the events. \n\nHere is the query I'd like you to extract events from: baseball player hits ball<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",
"sequel_raw_generation": "EXPLANATION: It makes sense to extract events that are direct consequences or actions related to a baseball player hitting a ball, as these events are likely to be visualized in a video of the query. The extracted events should be specific and relevant to the sport of baseball, considering the typical actions and outcomes that occur when a player hits a ball.\n\nEVENTS:\n1. Baseball player swinging bat\n2. Ball flying through the air\n3. Ball landing in the outfield\n4. Runner rounding bases\n5. Fielder catching the ball",
"video_URL": "https://youtube.com/watch?v=video7021",
"video_description": null,
"video_id": "video7021"
} | 16 | [
"A baseball player in a white uniform with the number 20 is at bat, preparing to swing. A catcher in red gear and an umpire are positioned behind the batter. In the background, players in red uniforms are seated in the dugout.",
"The baseball player in the white uniform with the number 20 has just completed his s... | video7021 | A baseball game is taking place between two teams, one wearing white uniforms and the other wearing red uniforms. The video begins with a player from the white team, number 20, at bat, swinging and following through with the motion. The catcher from the red team is crouched and ready to catch the ball, while the umpire observes the play. After the swing, the catcher stands up, suggesting he has caught the ball, and the player from the white team moves forward.
The scene then shifts to the outfield, where a player from the red team, number 3, is running to retrieve a ball. He continues to run across the outfield, eventually approaching the fence and attempting to catch the ball near the ground.
The video then cuts to a different scene, showing a player from a team wearing black and white uniforms, number 14, walking towards the dugout with teammates. They engage in conversation and approach the dugout, joined by other teammates. The scene suggests a break in the game or a transition between innings, with the players interacting and moving towards the dugout. The overall context is a baseball game with multiple teams and players involved in different aspects of the game. |
End of preview. Expand
in Data Studio
README.md exists but content is empty.
- Downloads last month
- 18
Size of downloaded dataset files:
4.08 MB
Size of the auto-converted Parquet files:
4.08 MB
Number of rows:
1,000