README.md

metadata

library_name: litert
pipeline_tag: image-classification
tags:
  - vision
  - image-classification
  - google
  - computer-vision
datasets:
  - imagenet-1k
model-index:
  - name: litert-community/squeezenet1_1
    results:
      - task:
          type: image-classification
          name: Image Classification
        dataset:
          name: ImageNet-1k
          type: imagenet-1k
          config: default
          split: validation
        metrics:
          - name: Top 1 Accuracy (Full Precision)
            type: accuracy
            value: 0.5819
          - name: Top 5 Accuracy (Full Precision)
            type: accuracy
            value: 0.8059
          - name: Top 1 Accuracy (Dynamic Quantized wi8 afp32)
            type: accuracy
            value: 0.5809
          - name: Top 5 Accuracy (Dynamic Quantized wi8 afp32)
            type: accuracy
            value: 0.8053

Squeezenet1_1

SqueezeNet 1.1 is a highly efficient model pre-trained on the ImageNet-1k dataset at a 224x224 resolution. Detailed in the paper "SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size" and released via its official repository, this updated version improves upon SqueezeNet 1.0. It reduces computational costs by 2.4x (operating at just 0.35 GFLOPS) and uses slightly fewer parameters, all while maintaining the exact same level of accuracy.

Model description

The model was converted from a checkpoint from PyTorch Vision (SqueezeNet1_1_Weights.IMAGENET1K_V1).

The original model has:
acc@1 (on ImageNet-1K): 58.178%
acc@5 (on ImageNet-1K): 80.624%
num_params: 1,235,496

This model is released under the BSD 3-Clause License, inheriting the license of the torchvision repository from which it was converted.

squeezenet1_1_int8_channelwise.tflite: Mixed INT8/FP32 with channelwise INT8 weights. Final pooling remains FP32; input is INT8 and output is FLOAT32.

Compatibility

File	CPU	GPU	NPU
squeezenet1_1.tflite	Supported	Supported	N/A
squeezenet1_1_int8_channelwise.tflite	Supported	Not supported	Qualcomm / MediaTek

Intended uses & limitations

The model files were converted from pretrained weights from PyTorch Vision. The models may have their own licenses or terms and conditions derived from PyTorch Vision and the dataset used for training. It is your responsibility to determine whether you have permission to use the models for your use case.

How to Use

​​1. Install Dependencies

Ensure your Python environment is set up with the required libraries. Run the following command in your terminal

pip install numpy Pillow huggingface_hub ai-edge-litert

2. Prepare Your Image

The script expects an image file to analyze. Make sure you have an image (e.g., cat.jpg or car.png) saved in the same working directory as your script.

3. Save the Script

Create a new file named classify.py, paste the script below into it, and save the file:

#!/usr/bin/env python3
import os
import argparse
import json
import numpy as np
from PIL import Image
from huggingface_hub import hf_hub_download
from ai_edge_litert.compiled_model import CompiledModel

def preprocess(img: Image.Image) -> np.ndarray:
  img = img.convert("RGB")
  w, h = img.size

  # Resize shortest edge to 256
  s = 256
  if w < h:
      img = img.resize((s, int(h * s / w)), Image.BILINEAR)
  else:
      img = img.resize((int(w * s / h), s), Image.BILINEAR)

  # Central crop to 224x224
  left = int(round((img.size[0] - 224) / 2.0))
  top = int(round((img.size[1] - 224) / 2.0))
  img = img.crop((left, top, left + 224, top + 224))

  # Rescale to [0.0, 1.0] and Normalize
  x = np.asarray(img, dtype=np.float32) / 255.0
  x = (x - np.array([0.485, 0.456, 0.406], dtype=np.float32)) / np.array(
      [0.229, 0.224, 0.225], dtype=np.float32
  )

  # Transpose from HWC (224, 224, 3) to CHW (3, 224, 224)
  x = np.transpose(x, (2, 0, 1))

  # Add the batch dimension to create NCHW (1, 3, 224, 224)
  x = np.expand_dims(x, axis=0)

  # The C++ buffer reads the bytes in the correct NCHW order.
  x = np.ascontiguousarray(x, dtype=np.float32)

  return x

def main():
  ap = argparse.ArgumentParser()
  ap.add_argument("--image", required=True, help="Path to the input image")
  args = ap.parse_args()

  # Download the TFLite model and labels for squeezenet1_1
  model_path = hf_hub_download("litert-community/squeezenet1_1", "squeezenet1_1.tflite")
  labels_path = hf_hub_download(
      "huggingface/label-files", "imagenet-1k-id2label.json", repo_type="dataset"
  )

  with open(labels_path, "r", encoding="utf-8") as f:
      id2label = {int(k): v for k, v in json.load(f).items()}

  img = Image.open(args.image)
  x = preprocess(img)

  model = CompiledModel.from_file(model_path)
  inp = model.create_input_buffers(0)
  out = model.create_output_buffers(0)

  # The 4D tensor is perfectly aligned with the C++ memory expectations
  inp[0].write(x)
  model.run_by_index(0, inp, out)

  req = model.get_output_buffer_requirements(0, 0)
  y = out[0].read(req["buffer_size"] // np.dtype(np.float32).itemsize, np.float32)

  pred = int(np.argmax(y))
  label = id2label.get(pred, f"class_{pred}")

  print(f"Top-1 class index: {pred}")
  print(f"Top-1 label: {label}")

if __name__ == "__main__":
  main()

4. Execute the Python Script

Run the below command

python classify.py --image cat.jpg

BibTeX Entry and Citation Info

@misc{iandola2016squeezenetalexnetlevelaccuracy50x,
      title={SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size}, 
      author={Forrest N. Iandola and Song Han and Matthew W. Moskewicz and Khalid Ashraf and William J. Dally and Kurt Keutzer},
      year={2016},
      eprint={1602.07360},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/1602.07360}, 
}