微调中文PPOCR-v4出错，输入数据维度不匹配

[0-Maxwei-0]

return self.forward(*inputs, **kwargs)

File "/root/miniconda3/lib/python3.8/site-packages/paddle/nn/layer/conv.py", line 712, in forward
out = F.conv._conv_nd(
File "/root/miniconda3/lib/python3.8/site-packages/paddle/nn/functional/conv.py", line 140, in _conv_nd
pre_bias = _C_ops.conv2d(
ValueError: (InvalidArgument) The input of Op(Conv) should be a 4-D or 5-D Tensor. But received: input's dimension is 3, input's shape is [64, 240, 256].
[Hint: Expected in_dims.size() == 4 || in_dims.size() == 5 == true, but received in_dims.size() == 4 || in_dims.size() == 5:0 != true:1.] (at /paddle/paddle/phi/infermeta/binary.cc:470)

[0-Maxwei-0]

同一批数据，在v3上调是没问题的，v4上就会出现这样的错误，不知道是什么原因呢？谢谢

[IbrarBabar009]

Please share your yml file

and what is the version you are using of PaddleOCR, I am asking about the version of github PaddleOCR you are using?

share a more output please

[0-Maxwei-0]

Please share your yml file

and what is the version you are using of PaddleOCR, I am asking about the version of github PaddleOCR you are using?

share a more output please

Global:
debug: false
use_gpu: true
epoch_num: 1000
log_smooth_window: 20
print_batch_step: 20
save_model_dir: ./output/rec_dkd_400w_svtr_ctc_lcnet_blank_dkd0.1/
save_epoch_step: 40
eval_batch_step:

• 0
• 2000
  cal_metric_during_train: true
  pretrained_model: /root/autodl-tmp/ch_PP-OCRv4_rec_train/student.pdparams
  checkpoints: # ./output/rec_dkd_400w_svtr_ctc_lcnet_blank_dkd0.1/latest
  save_inference_dir: null
  use_visualdl: false
  infer_img: doc/imgs_words/ch/word_1.jpg
  character_dict_path: ppocr/utils/ppocr_keys_v1.txt
  max_text_length: &max_text_length 40
  infer_mode: false
  use_space_char: true
  distributed: true
  save_res_path: ./output/rec/predicts_ppocrv3.txt
  Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
  name: Cosine
  learning_rate: 0.0001
  warmup_epoch: 2
  regularizer:
  name: L2
  factor: 3.0e-05
  Architecture:
  model_type: rec
  name: DistillationModel
  algorithm: Distillation
  Models:
  Teacher:
  pretrained:
  freeze_params: true
  return_all_feats: true
  model_type: rec
  algorithm: SVTR
  Transform: null
  Backbone:
  name: SVTRNet
  img_size:
  - 48
  - 320
  out_char_num: 40
  out_channels: 192
  patch_merging: Conv
  embed_dim:
  - 64
  - 128
  - 256
  depth:
  - 3
  - 6
  - 3
  num_heads:
  - 2
  - 4
  - 8
  mixer:
  - Conv
  - Conv
  - Conv
  - Conv
  - Conv
  - Conv
  - Global
  - Global
  - Global
  - Global
  - Global
  - Global
  local_mixer:
  - - 5
  - 5
  - - 5
  - 5
  - - 5
  - 5
  last_stage: false
  prenorm: true
  Head:
  name: MultiHead
  head_list:
  - CTCHead:
  Neck:
  name: svtr
  dims: 120
  depth: 2
  hidden_dims: 120
  kernel_size: [1, 3]
  use_guide: True
  Head:
  fc_decay: 0.00001
  - NRTRHead:
  nrtr_dim: 384
  max_text_length: *max_text_length
  Student:
  pretrained:
  freeze_params: false
  return_all_feats: true
  model_type: rec
  algorithm: SVTR
  Transform: null
  Backbone:
  name: PPLCNetV3
  scale: 0.95
  Head:
  name: MultiHead
  head_list:
  - CTCHead:
  Neck:
  name: svtr
  dims: 120
  depth: 2
  hidden_dims: 120
  kernel_size: [1, 3]
  use_guide: True
  Head:
  fc_decay: 0.00001
  - NRTRHead:
  nrtr_dim: 384
  max_text_length: *max_text_length
  Loss:
  name: CombinedLoss
  loss_config_list:
• DistillationDKDLoss:
  weight: 0.1
  model_name_pairs:
  • • Student
    • Teacher
      key: head_out
      multi_head: true
      alpha: 1.0
      beta: 2.0
      dis_head: gtc
      name: dkd
• DistillationCTCLoss:
  weight: 1.0
  model_name_list:
  • Student
    key: head_out
    multi_head: true
• DistillationNRTRLoss:
  weight: 1.0
  smoothing: false
  model_name_list:
  • Student
    key: head_out
    multi_head: true
• DistillCTCLogits:
  weight: 1.0
  reduction: mean
  model_name_pairs:
  • • Student
    • Teacher
      key: head_out
      PostProcess:
      name: DistillationCTCLabelDecode
      model_name:
• Student
  key: head_out
  multi_head: true
  Metric:
  name: DistillationMetric
  base_metric_name: RecMetric
  main_indicator: acc
  key: Student
  ignore_space: false
  Train:
  dataset:
  name: SimpleDataSet
  data_dir: /root/autodl-tmp/rec/train
  label_file_list:
  • /root/autodl-tmp/rec/train.txt
    ratio_list:
  • 1.0
    transforms:
  • DecodeImage:
    img_mode: BGR
    channel_first: false
  • RecAug:
  • MultiLabelEncode:
    gtc_encode: NRTRLabelEncode
  • RecResizeImg:
    image_shape: [3, 48, 320]
  • KeepKeys:
    keep_keys:
    • image
    • label_ctc
    • label_gtc
    • length
    • valid_ratio
      loader:
      shuffle: true
      batch_size_per_card: 64
      drop_last: true
      num_workers: 1
      use_shared_memory: true
      Eval:
      dataset:
      name: SimpleDataSet
      data_dir: /root/autodl-tmp/rec/val
      label_file_list:
  • /root/autodl-tmp/rec/val.txt
    transforms:
  • DecodeImage:
    img_mode: BGR
    channel_first: false
  • MultiLabelEncode:
    gtc_encode: NRTRLabelEncode
  • RecResizeImg:
    image_shape: [3, 48, 320]
  • KeepKeys:
    keep_keys:
    • image
    • label_ctc
    • label_gtc
    • length
    • valid_ratio
      loader:
      shuffle: false
      drop_last: false
      batch_size_per_card: 64
      num_workers: 1
      profiler_options: null

使用paddleOCR 2.6.1

[0-Maxwei-0]

我参考了这篇回答 #10700进行代码的修改。

[tohsakask]

不调整neck就行了，如果是SVTR还是用的reshape

[minakami443]

你用的config是Teacher-Student蒸餾用的, 會報錯正常
要fine-tune的話參照"\PaddleOCR\configs\rec\PP-OCRv4\ch_PP-OCRv4_rec.yml"去改

[iPengXPro]

请问v4模型微调您跑通了吗？能分指点一下如何操作吗？