检测训练模型微调问题

[kwanxian]

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

• 系统环境/System Environment：aistudio
• 版本号/Version：Paddle：2.5.1 PaddleOCR：2.7 问题相关组件/Related components：
• 运行指令/Command Code：python3 tools/train.py -c train_data/config/en_PP-OCRv3_det.yml
• 完整报错/Complete Error Message：训练约3小时， Precision、Recall、Hmean 均为0

我们提供了AceIssueSolver来帮助你解答问题，你是否想要它来解答(请填写yes/no)?/We provide AceIssueSolver to solve issues, do you want it? (Please write yes/no):

请尽量不要包含图片在问题中/Please try to not include the image in the issue.

训练集：1700
验证集：400

参考文档：模型微调
参考配置：ch_PP-OCRv3_det_student.yml
预训练模型：en_PP-OCRv3_det

训练配置如下：

Global:
  debug: false
  use_gpu: true
  epoch_num: 300
  log_smooth_window: 20
  print_batch_step: 100
  save_model_dir: ./output/det_ppocr_v3
  save_epoch_step: 300
  eval_batch_step: [0, 500]
  cal_metric_during_train: false
  # pretrained_model: https://paddleocr.bj.bcebos.com/pretrained/MobileNetV3_large_x0_5_pretrained.pdparams
  pretrained_model: ./train_data/model/en_PP-OCRv3_det_train/student.pdparams
  checkpoints: null
  save_inference_dir: ./inference/det_ppocr_v3/
  use_visualdl: false
  infer_img: doc/imgs_en/img_10.jpg
  save_res_path: ./output/det_ppocr_v3/predicts_ppocrv3.txt
  distributed: true

Architecture:
  model_type: det
  algorithm: DB
  Transform:
  Backbone:
    name: MobileNetV3
    scale: 0.5
    model_name: large
    disable_se: True
  Neck:
    name: RSEFPN
    out_channels: 96
    shortcut: True
  Head:
    name: DBHead
    k: 50

Loss:
  name: DBLoss
  balance_loss: true
  main_loss_type: DiceLoss
  alpha: 5
  beta: 10
  ohem_ratio: 3
Optimizer:
  name: Adam
  beta1: 0.9
  beta2: 0.999
  lr:
    name: Cosine
    learning_rate: 0.001
    warmup_epoch: 2
  regularizer:
    name: L2
    factor: 5.0e-05
PostProcess:
  name: DBPostProcess
  thresh: 0.3
  box_thresh: 0.6
  max_candidates: 1000
  unclip_ratio: 1.5
  # 支持fast和slow
  score_mode: slow
  # 分割结果进行膨胀
  use_dilation: false
Metric:
  name: DetMetric
  main_indicator: hmean
Train:
  dataset:
    name: SimpleDataSet
    data_dir: ./train_data/train/
    label_file_list:
      - ./train_data/train/Label.txt
    ratio_list: [1.0]
    transforms:
    - DecodeImage:
        img_mode: BGR
        channel_first: false
    - DetLabelEncode: null
    - IaaAugment:
        augmenter_args:
        - type: Fliplr
          args:
            p: 0.5
        - type: Affine
          args:
            rotate:
            - -10
            - 10
        - type: Resize
          args:
            size:
            - 0.5
            - 3
    - EastRandomCropData:
        size:
        - 960
        - 960
        max_tries: 50
        keep_ratio: true
    - MakeBorderMap:
        shrink_ratio: 0.4
        thresh_min: 0.3
        thresh_max: 0.7
    - MakeShrinkMap:
        shrink_ratio: 0.4
        min_text_size: 8
    - NormalizeImage:
        scale: 1./255.
        mean:
        - 0.485
        - 0.456
        - 0.406
        std:
        - 0.229
        - 0.224
        - 0.225
        order: hwc
    - ToCHWImage: null
    - KeepKeys:
        keep_keys:
        - image
        - threshold_map
        - threshold_mask
        - shrink_map
        - shrink_mask
  loader:
    shuffle: true
    drop_last: false
    batch_size_per_card: 8
    num_workers: 4
    use_shared_memory: false
Eval:
  dataset:
    name: SimpleDataSet
    data_dir: ./train_data/valid/
    label_file_list:
      - ./train_data/valid/Label.txt
    transforms:
    - DecodeImage:
        img_mode: BGR
        channel_first: false
    - DetLabelEncode: null
    - DetResizeForTest: 
        limit_side_len: 960
        limit_type: max
    - NormalizeImage:
        scale: 1./255.
        mean:
        - 0.485
        - 0.456
        - 0.406
        std:
        - 0.229
        - 0.224
        - 0.225
        order: hwc
    - ToCHWImage: null
    - KeepKeys:
        keep_keys:
        - image
        - shape
        - polys
        - ignore_tags
  loader:
    shuffle: false
    drop_last: false
    batch_size_per_card: 1
    num_workers: 2
    use_shared_memory: false

跑完warmup_epoch后，Precision、Recall、Hmean 一直打印为0

[2023/09/12 11:19:07] ppocr INFO: epoch: [29/300], global_step: 6000, lr: 0.000981, loss: 0.619097, loss_shrink_maps: 0.295784, loss_threshold_maps: 0.265661, loss_binary_maps: 0.059084, loss_cbn: 0.000000, avg_reader_cost: 0.87671 s, avg_batch_cost: 1.02437 s, avg_samples: 5.12, ips: 4.99817 samples/s, eta: 23:50:37
[2023/09/12 11:19:19] ppocr INFO: cur metric, precision: 0, recall: 0.0, hmean: 0, fps: 41.68181944490627
[2023/09/12 11:19:19] ppocr INFO: best metric, hmean: 0.16666666666666666, is_float16: False, precision: 0.303125, recall: 0.11492890995260663, fps: 38.38100947037132, best_epoch: 3
[2023/09/12 11:21:45] ppocr INFO: epoch: [29/300], global_step: 6100, lr: 0.000981, loss: 0.603158, loss_shrink_maps: 0.276820, loss_threshold_maps: 0.261379, loss_binary_maps: 0.055386, loss_cbn: 0.000000, avg_reader_cost: 1.20307 s, avg_batch_cost: 1.43353 s, avg_samples: 8.0, ips: 5.58061 samples/s, eta: 23:47:15

[cc12135]

改一下eastrandomcropdata