AblationStudy look for help .looking for FRM,MSR,SGM where

[YMT-ymt]

Hello, may I ask if I need to conduct ablation experiments on the FRM, SGM, and MSR modules in svtrv2.

1. For MSR, I changed RatioDataSetTVResize to RatioDataSet in the configuration file (svtrv2_stmtr_gtc_rctc_ch. yml). Does this have the effect of removing MSR for ablation experiments.
2. For FRM, change SVTRv2LNConvTwo33 to the original SVTRv2 to remove FRM for ablation experiments.
3. For SGM, I haven't found where to use it, so I can't conduct ablation experiments
   Did I correctly identify the innovative module in your paper and conduct the ablation experiment in this way. Can you tell me where these three ablation experiment modules are located? In which specific folder?
   I hope you can help me. Thank you very much for watching my question. If possible, I would like to request an answer. Thank you very much.

[t1083714]

I also want to know, hope to answer, thank you very much.

[Topdu]

without SGM: changing GTCDecoder to RCTCDecoder, refer to https://github.com/Topdu/OpenOCR/blob/main/configs/rec/svtrv2/svtrv2_rctc.yml,

svtrv2 without MSR and without FRM
without MSR: changing RatioDataSetTVResize to LMDBDataSet
without FRM: changing RCTCDecoder to CTCDecoder

Global:
  device: gpu
  epoch_num: 20
  log_smooth_window: 20
  print_batch_step: 10
  output_dir: ./output/rec/u14m_filter/svtrv2_ctc_wo_msr/
  save_epoch_step: [15, 1]
  # evaluation is run every 2000 iterations
  eval_epoch_step: [0, 1]
  eval_batch_step: [0, 500]
  cal_metric_during_train: True
  pretrained_model:
  checkpoints:
  use_tensorboard: false
  infer_img:
  # for data or label process
  character_dict_path: &character_dict_path ./tools/utils/EN_symbol_dict.txt
  max_text_length: &max_text_length 25
  use_space_char: &use_space_char False
  save_res_path: ./output/rec/u14m_filter/predicts_svtrv2_ctc_wo_msr.txt
  use_amp: True

Optimizer:
  name: AdamW
  lr: 0.00065 # for 4gpus bs256/gpu
  weight_decay: 0.05
  filter_bias_and_bn: True

LRScheduler:
  name: OneCycleLR
  warmup_epoch: 1.5 # pct_start 0.075*20 = 1.5ep
  cycle_momentum: False

Architecture:
  model_type: rec
  algorithm: SVTRv2
  Transform:
  Encoder:
    name: SVTRv2LNConvTwo33
    use_pos_embed: False
    out_channels: 256
    dims: [128, 256, 384]
    depths: [6, 6, 6]
    num_heads: [4, 8, 12]
    mixer: [['Conv','Conv','Conv','Conv','Conv','Conv'],['Conv','Conv','FGlobal','Global','Global','Global'],['Global','Global','Global','Global','Global','Global']]
    local_k: [[5, 5], [5, 5], [-1, -1]]
    sub_k: [[1, 1], [2, 1], [-1, -1]]
    last_stage: True
    feat2d: False
  Decoder:
    name: CTCDecoder

Loss:
  name: CTCLoss
  zero_infinity: True

PostProcess:
  name: CTCLabelDecode

Metric:
  name: RecMetric
  main_indicator: acc
  is_filter: True

Train:
  dataset:
    name: LMDBDataSet
    data_dir: ../Union14M-L-LMDB-Filtered
    transforms:
      - DecodeImagePIL: # load image
          img_mode: RGB
      - PARSeqAugPIL:
      - 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: [32, 128]
          padding: False
      - KeepKeys:
          keep_keys: ['image', 'label', 'length'] # dataloader will return list in this order
  loader:
    shuffle: True
    batch_size_per_card: 256
    drop_last: True
    num_workers: 4

Eval:
  dataset:
    name: LMDBDataSet
    data_dir: ../evaluation/
    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: [32, 128]
          padding: False
      - 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: 2