Why evaluation score always 0.0

[minhhoang2705]

I have been training CRNN for Vietnamese text recognition for days but my evaluation score always 0.0 in spite of my training score improve

Here is my CRNN config file. If there anything wrong please let me know.

Global:
  use_gpu: true
  epoch_num: 500
  log_smooth_window: 20
  print_batch_step: 10
  save_model_dir: ./output/rec/r34_vd_none_bilstm_ctc_vie_v2/
  save_epoch_step: 5
  # evaluation is run every 2000 iterations
  eval_batch_step: [0, 500]
  cal_metric_during_train: True
  pretrained_model: ./pretrain_models/rec_r34_vd_none_bilstm_ctc_v2.0_train/best_accuracy
  checkpoints: ./output/rec/r34_vd_none_bilstm_ctc_vie_v2/latest
  save_inference_dir:
  use_visualdl: False
  infer_img: null
  # for data or label process
  character_dict_path: ./ppocr/utils/dict/custom.txt
  max_text_length: 1000
  infer_mode: False
  use_space_char: True
  save_res_path: ./output/rec/predicts_r34_vd_none_bilstm_ctc.txt

Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
    learning_rate: 0.0005
  regularizer:
    name: 'L2'
    factor: 0

Architecture:
  model_type: rec
  algorithm: CRNN
  Transform:
  Backbone:
    name: ResNet
    layers: 34
  Neck:
    name: SequenceEncoder
    encoder_type: rnn
    hidden_size: 256
  Head:
    name: CTCHead
    fc_decay: 0

Loss:
  name: CTCLoss

PostProcess:
  name: CTCLabelDecode

Metric:
  name: RecMetric
  main_indicator: acc

Train:
  dataset:
    name: SimpleDataSet
    data_dir: ./train_data/2nd_phase/rec_data/double_word_data
    label_file_list: ./train_data/2nd_phase/rec_data/train_data.txt
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - CTCLabelEncode: # Class handling label
      - RecResizeImg:
          image_shape: [3, 32, 1024]
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: True
    batch_size_per_card: 32
    drop_last: True
    num_workers: 8

Eval:
  dataset:
    name: SimpleDataSet
    data_dir: ./train_data/2nd_phase/rec_data/Vietnamese
    label_file_list: ./train_data/2nd_phase/rec_data/output.txt
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - CTCLabelEncode: # Class handling label
      - RecResizeImg:
          image_shape: [3, 32, 1024]
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: False
    drop_last: False
    batch_size_per_card: 16
    num_workers: 6

[VioletEqz]

From the given logs, seems like your model are overfitted because of too few training data. You can try the following:

• Increasing the amount of training data.
• Add in some augmentation.

Reasonable amount of data should be somewhat around 50000 images, depending on the char dict you used.