rec_mtb_nrtr.yml

Global:
  use_gpu: True
  epoch_num: 21
  log_smooth_window: 20
  print_batch_step: 10
  save_model_dir: ./output/rec/nrtr/
  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: doc/imgs_words_en/word_10.png
  # for data or label process
  character_dict_path: 
  character_type: EN_symbol
  max_text_length: 25
  infer_mode: False
  use_space_char: True
  save_res_path: ./output/rec/predicts_nrtr.txt

Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.99
  clip_norm: 5.0
  lr:
    name: Cosine
    learning_rate: 0.0005
    warmup_epoch: 2
  regularizer:
    name: 'L2'
    factor: 0.

Architecture:
  model_type: rec
  algorithm: NRTR
  in_channels: 1
  Transform:
  Backbone:
    name: MTB
    cnn_num: 2
  Head:
    name: Transformer
    d_model: 512
    num_encoder_layers: 6
    beam_size: -1 # When Beam size is greater than 0, it means to use beam search when evaluation.
    

Loss:
  name: NRTRLoss
  smoothing: True

PostProcess:
  name: NRTRLabelDecode

Metric:
  name: RecMetric
  main_indicator: acc

Train:
  dataset:
    name: LMDBDataSet
    data_dir: ./train_data/data_lmdb_release/training/
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - NRTRLabelEncode: # Class handling label
      - NRTRRecResizeImg:
          image_shape: [100, 32]
          resize_type: PIL # PIL or OpenCV
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: True
    batch_size_per_card: 512
    drop_last: True
    num_workers: 8

Eval:
  dataset:
    name: LMDBDataSet
    data_dir: ./train_data/data_lmdb_release/evaluation/
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
      - NRTRLabelEncode: # Class handling label
      - NRTRRecResizeImg:
          image_shape: [100, 32]
          resize_type: PIL # PIL or OpenCV
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: False
    drop_last: False
    batch_size_per_card: 256
    num_workers: 1
    use_shared_memory: False