e2e_r50_vd_pg.yml 3.21 KB
Newer Older
Jethong's avatar
Jethong committed
1
Global:
2
  use_gpu: True
Jethong's avatar
Jethong committed
3
4
  epoch_num: 600
  log_smooth_window: 20
5
  print_batch_step: 10
Jethong's avatar
Jethong committed
6
  save_model_dir: ./output/pgnet_r50_vd_totaltext/
7
8
  save_epoch_step: 10
  # evaluation is run every 0 iterationss after the 1000th iteration
Jethong's avatar
Jethong committed
9
  eval_batch_step: [ 0, 1000 ]
10
11
12
13
  # 1. If pretrained_model is saved in static mode, such as classification pretrained model
  #    from static branch, load_static_weights must be set as True.
  # 2. If you want to finetune the pretrained models we provide in the docs,
  #    you should set load_static_weights as False.
Jethong's avatar
Jethong committed
14
  load_static_weights: False
Jethong's avatar
Jethong committed
15
16
17
18
19
20
  cal_metric_during_train: False
  pretrained_model:
  checkpoints:
  save_inference_dir:
  use_visualdl: False
  infer_img:
Jethong's avatar
Jethong committed
21
  valid_set: totaltext # two mode: totaltext valid curved words, partvgg valid non-curved words
Jethong's avatar
Jethong committed
22
  save_res_path: ./output/pgnet_r50_vd_totaltext/predicts_pgnet.txt
Jethong's avatar
Jethong committed
23
  character_dict_path: ppocr/utils/ic15_dict.txt
Jethong's avatar
Jethong committed
24
  character_type: EN
Jethong's avatar
Jethong committed
25
26
27
  max_text_length: 50 # the max length in seq
  max_text_nums: 30 # the max seq nums in a pic
  tcl_len: 64
Jethong's avatar
Jethong committed
28
29
30

Architecture:
  model_type: e2e
31
  algorithm: PGNet
Jethong's avatar
Jethong committed
32
33
34
35
36
37
38
39
40
41
42
  Transform:
  Backbone:
    name: ResNet
    layers: 50
  Neck:
    name: PGFPN
  Head:
    name: PGHead

Loss:
  name: PGLoss
Jethong's avatar
Jethong committed
43
44
45
46
  tcl_bs: 64
  max_text_length: 50 # the same as Global: max_text_length
  max_text_nums: 30 # the same as Global:max_text_nums
  pad_num: 36 # the length of dict for pad
Jethong's avatar
Jethong committed
47
48

Optimizer:
49
50
51
  name: Adam
  beta1: 0.9
  beta2: 0.999
Jethong's avatar
Jethong committed
52
53
54
55
  lr:
    learning_rate: 0.001
  regularizer:
    name: 'L2'
56
57
    factor: 0

Jethong's avatar
Jethong committed
58
59
60

PostProcess:
  name: PGPostProcess
Jethong's avatar
Jethong committed
61
  score_thresh: 0.5
Jethong's avatar
Jethong committed
62
  mode: fast   # fast or slow two ways
Jethong's avatar
Jethong committed
63

Jethong's avatar
Jethong committed
64
65
Metric:
  name: E2EMetric
Jethong's avatar
Jethong committed
66
  mode: A   # A or B
Jethong's avatar
fix yml  
Jethong committed
67
  gt_mat_dir:  ./train_data/total_text/gt  # the dir of gt_mat
Jethong's avatar
Jethong committed
68
  character_dict_path: ppocr/utils/ic15_dict.txt
Jethong's avatar
Jethong committed
69
  main_indicator: f_score_e2e
Jethong's avatar
Jethong committed
70
71
72

Train:
  dataset:
Jethong's avatar
Jethong committed
73
    name: PGDataSet
Jethong's avatar
fix yml  
Jethong committed
74
75
    data_dir: ./train_data/total_text/train
    label_file_list: [./train_data/total_text/train/]
76
    ratio_list: [1.0]
Jethong's avatar
Jethong committed
77
78
79
80
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
Jethong's avatar
Jethong committed
81
      - E2ELabelEncode_train:
Jethong's avatar
Jethong committed
82
      - PGProcessTrain:
Jethong's avatar
Jethong committed
83
          batch_size: 14  # same as loader: batch_size_per_card
Jethong's avatar
Jethong committed
84
85
86
87
88
89
90
91
          min_crop_size: 24
          min_text_size: 4
          max_text_size: 512
      - KeepKeys:
          keep_keys: [ 'images', 'tcl_maps', 'tcl_label_maps', 'border_maps','direction_maps', 'training_masks', 'label_list', 'pos_list', 'pos_mask' ] # dataloader will return list in this order
  loader:
    shuffle: True
    drop_last: True
92
93
    batch_size_per_card: 14
    num_workers: 16
Jethong's avatar
Jethong committed
94
95
96

Eval:
  dataset:
97
    name: PGDataSet
Jethong's avatar
fix yml  
Jethong committed
98
    data_dir: ./train_data/total_text/test
99
    label_file_list: [./train_data/total_text/test/]
Jethong's avatar
Jethong committed
100
101
    transforms:
      - DecodeImage: # load image
Jethong's avatar
Jethong committed
102
          img_mode: RGB
Jethong's avatar
Jethong committed
103
          channel_first: False
Jethong's avatar
Jethong committed
104
      - E2ELabelEncode_test:
Jethong's avatar
Jethong committed
105
106
107
108
109
110
111
112
113
      - E2EResizeForTest:
          max_side_len: 768
      - NormalizeImage:
          scale: 1./255.
          mean: [ 0.485, 0.456, 0.406 ]
          std: [ 0.229, 0.224, 0.225 ]
          order: 'hwc'
      - ToCHWImage:
      - KeepKeys:
Jethong's avatar
Jethong committed
114
          keep_keys: [ 'image', 'shape', 'polys', 'texts', 'tags', 'img_id']
Jethong's avatar
Jethong committed
115
116
117
118
119
  loader:
    shuffle: False
    drop_last: False
    batch_size_per_card: 1 # must be 1
    num_workers: 2