微调训练ch_PP-OCRv3_det_distill_train的student模型，无论如何训练改参数，eval的准确率不超过50%，而且曾经能预测也无法预测

[zxy2020]

数据集大小：1600（train）+ 500 (val) 数据集制作方式： 使用paddleocr 推理后选出没有问题导入，有问题的重新标记，但是训练后原来能检测也不能检测了 数据集内容： GPU: Teslav100 *4 32G （aistudio），单卡也训过，（单卡8batch lr 0.0001, 4卡 0.0005） 配置文件：configs/det/ch_PP-OCRv3/ch_PP-OCRv3_det_student.yml

开始训练前指标： 训练后的指标：

Global: debug: false use_gpu: true epoch_num: 200 log_smooth_window: 20 print_batch_step: 10 save_model_dir: ./output/0712_ch_PP-OCR_V3_det/ save_epoch_step: 100 eval_batch_step:

• 0
• 50 cal_metric_during_train: false pretrained_model: /home/aistudio/PaddleOCR/ch_PP-OCRv3_det_distill_train/student.pdparams checkpoints: /home/aistudio/PaddleOCR/output/0712_ch_PP-OCR_V3_det/latest.pdparams save_inference_dir: null use_visualdl: false infer_img: doc/imgs_en/img_10.jpg save_res_path: ./checkpoints/det_db/predicts_db.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: 0.0005 # 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: 2 Metric: name: DetMetric main_indicator: hmean Train: dataset: name: SimpleDataSet data_dir: /home/aistudio/art_det_good_2800/train label_file_list:

• /home/aistudio/art_det_good_2800/train.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 Eval: dataset: name: SimpleDataSet data_dir: /home/aistudio/23_0609_yk_art_500/train label_file_list:
• /home/aistudio/23_0609_yk_art_500/train.txt transforms:
  • DecodeImage: img_mode: BGR channel_first: false
  • DetLabelEncode: null
  • DetResizeForTest: null
  • 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

[izhaolinger]

是否是epoch少了？

[zxy2020]

是否是epoch少了？ 那我再试一试原始的轮速 你好，我只有2000张图片微调的，大概100epoch。 数据量和epoch的轮数有关系吗？

[izhaolinger]

是否是epoch少了？ 那我再试一试原始的轮速 你好，我只有2000张图片微调的，大概100epoch。 数据量和epoch的轮数有关系吗？

我和你一样，用的ch_PP-OCRv3_det_distill_train的student做微调，数据量上和你也差不多，1000个epoch，hmean:0.75，看到了一些效果

[zxy2020]

我和你一样，用的ch_PP-OCRv3_det_distill_train的student做微调，数据量上和你也差不多，1000个epoch，hmean:0.75，看到了一些效果 请问可以看下你的训练的配置文件吗？ 是否还修改了其他参数，GPU数，学习率调节，因为我的数据单一，我把数据增强都关了

[izhaolinger]

我和你一样，用的ch_PP-OCRv3_det_distill_train的student做微调，数据量上和你也差不多，1000个epoch，hmean:0.75，看到了一些效果 请问可以看下你的训练的配置文件吗？ 是否还修改了其他参数，GPU数，学习率调节，因为我的数据单一，我把数据增强都关了

可以的，这是我的配置文件：

Global: debug: false use_gpu: true epoch_num: 1000 log_smooth_window: 20 print_batch_step: 10 save_model_dir: ./output/ch_PP-OCR_V3_det/ save_epoch_step: 100 eval_batch_step:

• 0
• 400 cal_metric_during_train: false pretrained_model: /data/project/PaddleOCR-release-2.6/model/ch_PP-OCRv3_det_distill_train/student.pdparams checkpoints: null save_inference_dir: null use_visualdl: false infer_img: doc/imgs_en/img_10.jpg save_res_path: ./checkpoints/det_db/predicts_db.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: 0.0001 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 Metric: name: DetMetric main_indicator: hmean Train: dataset: name: SimpleDataSet data_dir: /data/project/PaddleOCR-release-2.6/train_data/ label_file_list: [ '/data/project/PaddleOCR-release-2.6/train_data/1/det/train.txt', '/data/project/PaddleOCR-release-2.6/train_data/2/det/train.txt', '/data/project/PaddleOCR-release-2.6/train_data/3/det/train.txt' ] ratio_list: [ 1.0, 1.0, 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: 0 Eval: dataset: name: SimpleDataSet data_dir: /data/project/PaddleOCR-release-2.6/train_data/ label_file_list: [ '/data/project/PaddleOCR-release-2.6/train_data/1/det/val.txt', '/data/project/PaddleOCR-release-2.6/train_data/2/det/val.txt', '/data/project/PaddleOCR-release-2.6/train_data/3/det/val.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: 0

GPU的话，我用两个两张GPU。

[UserWangZz]

该issue长时间未更新，暂将此issue关闭，如有需要可重新开启。