rec_r31_robustscanner.yml 2.6 KB
Newer Older
xuyang2233's avatar
xuyang2233 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Global:
  use_gpu: true
  epoch_num: 5
  log_smooth_window: 20
  print_batch_step: 20
  save_model_dir: ./output/rec/rec_r31_robustscanner/
  save_epoch_step: 1
  # evaluation is run every 2000 iterations
  eval_batch_step: [0, 2000]
  cal_metric_during_train: True
  pretrained_model:
  checkpoints: 
  save_inference_dir:
  use_visualdl: False
  infer_img: ./inference/rec_inference
  # for data or label process
  character_dict_path: ppocr/utils/dict90.txt
xuyang2233's avatar
xuyang2233 已提交
18
  max_text_length: &max_text_length 40
xuyang2233's avatar
xuyang2233 已提交
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
  infer_mode: False
  use_space_char: False
  rm_symbol: True
  save_res_path: ./output/rec/predicts_robustscanner.txt

Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
    name: Piecewise
    decay_epochs: [3, 4]
    values: [0.001, 0.0001, 0.00001] 
  regularizer:
    name: 'L2'
    factor: 0

Architecture:
  model_type: rec
  algorithm: RobustScanner
  Transform:
  Backbone:
xuyang2233's avatar
xuyang2233 已提交
41
    name: ResNet31
xuyang2233's avatar
xuyang2233 已提交
42 43 44 45 46 47 48 49 50 51
  Head:
    name: RobustScannerHead
    enc_outchannles: 128
    hybrid_dec_rnn_layers: 2
    hybrid_dec_dropout: 0
    position_dec_rnn_layers: 2
    start_idx: 91
    mask: True
    padding_idx: 92
    encode_value: False
xuyang2233's avatar
xuyang2233 已提交
52
    max_text_length: *max_text_length
xuyang2233's avatar
xuyang2233 已提交
53 54 55 56 57 58 59 60 61 62 63 64 65 66

Loss:
  name: SARLoss

PostProcess:
  name: SARLabelDecode

Metric:
  name: RecMetric
  is_filter: True


Train:
  dataset:
xuyang2233's avatar
xuyang2233 已提交
67 68 69
    name: SimpleDataSet
    label_file_list: ['./train_data/train_list.txt']
    data_dir: ./train_data/
xuyang2233's avatar
xuyang2233 已提交
70 71 72 73 74 75 76 77
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - SARLabelEncode: # Class handling label
      - RobustScannerRecResizeImg:
          image_shape: [3, 48, 48, 160] # h:48 w:[48,160]
          width_downsample_ratio: 0.25
xuyang2233's avatar
xuyang2233 已提交
78
          max_text_length: *max_text_length
xuyang2233's avatar
xuyang2233 已提交
79 80 81 82
      - KeepKeys:
          keep_keys: ['image', 'label', 'valid_ratio', 'word_positons'] # dataloader will return list in this order
  loader:
    shuffle: True
xuyang2233's avatar
xuyang2233 已提交
83
    batch_size_per_card: 64
xuyang2233's avatar
xuyang2233 已提交
84
    drop_last: True
xuyang2233's avatar
xuyang2233 已提交
85
    num_workers: 8
xuyang2233's avatar
xuyang2233 已提交
86 87 88 89 90
    use_shared_memory: False

Eval:
  dataset:
    name: LMDBDataSet
xuyang2233's avatar
xuyang2233 已提交
91
    data_dir: ./train_data/data_lmdb_release/evaluation/
xuyang2233's avatar
xuyang2233 已提交
92 93 94 95 96 97 98
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - SARLabelEncode: # Class handling label
      - RobustScannerRecResizeImg:
          image_shape: [3, 48, 48, 160]
xuyang2233's avatar
xuyang2233 已提交
99
          max_seq_len: *max_text_length
xuyang2233's avatar
xuyang2233 已提交
100 101 102 103 104 105
          width_downsample_ratio: 0.25
      - KeepKeys:
          keep_keys: ['image', 'label', 'valid_ratio', 'word_positons'] # dataloader will return list in this order
  loader:
    shuffle: False
    drop_last: False
xuyang2233's avatar
xuyang2233 已提交
106 107
    batch_size_per_card: 64
    num_workers: 4
xuyang2233's avatar
xuyang2233 已提交
108 109
    use_shared_memory: False