无法复现教程中的OCR v3检测模型的量化精度

[TinyQi]

我使用文档中：https://github.com/PaddlePaddle/PaddleSlim/tree/release/2.6/example/auto_compression/ocr的量化脚本进行量化，经过训练后模型的精度下降很大，请问我哪里做错了嘛？

精度的前后对比如下：

我使用的配置文件如下： Global: model_type: det model_dir: /share/disk3/xcq/02.model_cache/pretrain_models/ch_PP-OCRv3_det_infer_bak/ model_filename: inference.pdmodel params_filename: inference.pdiparams algorithm: DB

Distillation: alpha: 1.0 loss: l2

QuantAware: use_pact: true activation_bits: 8 is_full_quantize: false onnx_format: True activation_quantize_type: moving_average_abs_max weight_quantize_type: channel_wise_abs_max not_quant_pattern:

• skip_quant quantize_op_types:
• conv2d
• depthwise_conv2d weight_bits: 8

TrainConfig: epochs: 3 eval_iter: 200 learning_rate: type: CosineAnnealingDecay learning_rate: 0.00005 optimizer_builder: optimizer: type: Adam weight_decay: 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: /share/disk3/xcq/01.ImageData/010.OpenSourceData/icdar2015/text_localization/ label_file_list:

• /share/disk3/xcq/01.ImageData/010.OpenSourceData/icdar2015/text_localization/train_icdar2015_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: 4 num_workers: 0

Eval: dataset: name: SimpleDataSet data_dir: /share/disk3/xcq/01.ImageData/010.OpenSourceData/icdar2015/text_localization/ label_file_list:

• /share/disk3/xcq/01.ImageData/010.OpenSourceData/icdar2015/text_localization/test_icdar2015_label.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: 0

[ceci3]

我看上一个issue的log看起来是量化训练，可以参考 https://github.com/PaddlePaddle/PaddleSlim/blob/develop/example/post_training_quantization/detection/post_quant.py#L75 这个示例使用离线量化看下效果