detection_config.yml

Global:
  use_gpu: true
  use_xpu: false
  use_mlu: false
  epoch_num: 100
  log_smooth_window: 20
  print_batch_step: 10
  save_model_dir: output/detection
  save_epoch_step: 1000
  # evaluation is run every 2000 iterations
  eval_batch_step: [0, 2000]
  cal_metric_during_train: False
  pretrained_model: pretrained_models/detection/MobileNetV3_large_x0_5_pretrained.pdparams
  checkpoints:
  save_inference_dir:
  use_visualdl: False
  infer_img: 
  save_res_path: 

Architecture:
  model_type: det
  algorithm: DB
  Transform:
  Backbone:
    name: MobileNetV3
    scale: 0.5
    model_name: large
  Neck:
    name: DBFPN
    out_channels: 256
  Head:
    name: DBHead
    k: 50

Loss:
  name: DBLoss
  balance_loss: true
  main_loss_type: DiceLoss
  alpha: 5
  beta: 10
  ohem_ratio: 3

Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
    name: Cosine
    learning_rate: 0.001 # learning_rate
    warmup_epoch: 2
  regularizer:
    name: 'L2'
    factor: 0

PostProcess:
  name: DBPostProcess
  thresh: 0.3
  box_thresh: 0.6
  max_candidates: 1000
  unclip_ratio: 1.5

Metric:
  name: DetMetric
  main_indicator: hmean

Train:
  dataset:
    name: SimpleDataSet
    data_dir: ./
    label_file_list:
      - dataset/detection_train.txt
    ratio_list: [1.0]
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - DetLabelEncode: # Class handling label
      - IaaAugment:
          augmenter_args:
            - { 'type': Fliplr, 'args': { 'p': 0.5 } }
            # --- Augmentasi Ditambahkan (Menggunakan Nama Albumentations) ---
            - { 'type': GaussNoise, 'args': {'var_limit': [0, 165], 'p': 0.3 } } # Noise (prob 30%) - var_limit approx (0.05*255)^2
            - { 'type': GaussianBlur, 'args': { 'sigma_limit': [0.0, 1.5], 'p': 0.3 } } # Blur (prob 30%)
            - { 'type': RandomBrightnessContrast, 'args': { 'brightness_limit': 0.3, 'contrast_limit': 0.3, 'p': 0.5 } } # Light/Contrast (prob 50%)
            # --- Akhir Augmentasi Tambahan ---
            - { 'type': Affine, 'args': { 'rotate': [-10, 10] } } # Rotasi tetap ada
            - { 'type': Resize, 'args': { 'size': [0.5, 3] } } # Resize tetap ada
      - EastRandomCropData:
          size: [640, 640]
          max_tries: 50
          keep_ratio: true
      - MakeBorderMap:
          shrink_ratio: 0.4
          thresh_min: 0.3
          thresh_max: 0.7
      - MakeShrinkMap:
          shrink_ratio: 0.4
          min_text_size: 8
      - NormalizeImage:
          scale: 1./255.
          mean: [0.485, 0.456, 0.406]
          std: [0.229, 0.224, 0.225]
          order: 'hwc'
      - ToCHWImage:
      - KeepKeys:
          keep_keys: ['image', 'threshold_map', 'threshold_mask', 'shrink_map', 'shrink_mask'] # the order of the dataloader list
  loader:
    shuffle: True
    drop_last: False
    batch_size_per_card: 8
    num_workers: 4
    use_shared_memory: True

Eval:
  dataset:
    name: SimpleDataSet
    data_dir: ./
    label_file_list:
      - dataset/detection_test.txt
    transforms:
      - DecodeImage:
          img_mode: BGR
          channel_first: False
      - DetLabelEncode:
      - DetResizeForTest:
          image_shape: [736, 1280]
      - NormalizeImage:
          scale: 1./255.
          mean: [0.485, 0.456, 0.406]
          std: [0.229, 0.224, 0.225]
          order: 'hwc'
      - ToCHWImage:
      - KeepKeys:
          keep_keys: ['image', 'shape', 'polys', 'ignore_tags']
  loader:
    shuffle: False
    drop_last: False
    batch_size_per_card: 1 # must be 1 for detection eval
    num_workers: 4
    use_shared_memory: True