Zero accuracy never increases

[Santiagocanozz]

I've been trying to train a model with SVTRV2, but the accuracy always stays at zero. I've tweaked many parameters, but it still doesn't work. The idea is to use this baseline since it yields the best results.

Example config:
It was modified to load via CSV, and everything loaded successfully.

Global:
  device: gpu
  epoch_num: 100
  log_smooth_window: 20
  print_batch_step: 10
  output_dir: ./output/rec/ch/repsvtr_ch/
  save_epoch_step: [150, 10]
  # evaluation is run every 2000 iterations
  eval_epoch_step: [0, 1]
  eval_batch_step: [0, 2000]
  cal_metric_during_train: True
  pretrained_model: null
  checkpoints:
  use_tensorboard: false
  infer_img:
  # for data or label process
  character_dict_path: &character_dict_path /home/usrapp/OpenOCR/tools/utils/custom_dict.txt
  max_text_length: &max_text_length 10
  use_space_char: &use_space_char False
  save_res_path: ./output/rec/u14m_filter/predicts_svtrv2_ctc.txt
  use_amp: False
  project_name: resvtr_ctc_nosgm_ds

Optimizer:
  name: Adam
  lr: 0.00005
  weight_decay: 3.0e-05
  filter_bias_and_bn: False

LRScheduler:
  name: CosineAnnealingLR
  warmup_epoch: 5

Architecture:
  model_type: rec
  algorithm: SVTRv2_mobile
  Transform:
  Encoder:
    name: RepSVTREncoder
  Decoder:
    name: CTCDecoder
    svtr_encoder:
      dims: 256
      depth: 2
      hidden_dims: 256
      kernel_size: [1, 3]
      use_guide: True

Loss:
  name: CTCLoss
  zero_infinity: True

PostProcess:
  name: CTCLabelDecode
  character_dict_path: *character_dict_path

Metric:
  name: RecMetric
  main_indicator: acc
  ignore_space: True
  is_filter: True

Train:
  dataset:
    name: SimpleDataSet
    data_dir: /home/usrapp/qalpr-ml-data-ocr/data/Dataset_OCR/train
    label_file_list:
      - /home/usrapp/qalpr-ml-data-ocr/data/Dataset_OCR/Dataset_OCR_train.csv
    transforms:
      - DecodeImagePIL: # load image
          img_mode: RGB
      - CTCLabelEncode: # Class handling label
          character_dict_path: *character_dict_path
          use_space_char: *use_space_char
          max_text_length: *max_text_length
      - RecTVResize:
          image_shape: [48, 160]
          padding: False
      - KeepKeys:
          keep_keys: ['image', 'label', 'length']
  loader:
    shuffle: True
    batch_size_per_card: 32
    drop_last: True
    num_workers: 4

Eval:
  dataset:
    name: SimpleDataSet
    data_dir: /home/usrapp/qalpr-ml-data-ocr/data/Dataset_OCR/valid
    label_file_list:
        - /home/usrapp/qalpr-ml-data-ocr/data/Dataset_OCR_/Dataset_valid.csv
    transforms:
      - DecodeImage: # load image
          img_mode: RGB
      - CTCLabelEncode: # Class handling label
          character_dict_path: *character_dict_path
          use_space_char: *use_space_char
          max_text_length: *max_text_length
      - RecTVResize:
          image_shape: [48, 160]
          padding: False
      # - SVTRResize:
      #     image_shape: [3, 48, 160]
      #     padding: True
      - KeepKeys:
          keep_keys: ['image', 'label', 'length']
    loader:
      shuffle: False
      drop_last: False
      batch_size_per_card: 1
      num_workers: 4

[Topdu]

Please provide the full log.

[nvriese1]

Same issue as above