PaddleOCR 训练时遇到的问题

TM 2.0

已于 2024-04-16 15:36:36 修改

阅读量538

点赞数 3

文章标签：计算机视觉

于 2024-04-16 15:29:39 首次发布

本文链接：https://blog.csdn.net/qq_42693842/article/details/137826621

版权

PaddleOCR 训练时遇到的问题

config path: PaddleOCR/configs/det/ch_PP-OCRv4/ch_PP-OCRv4_det_cml.yml

PPLCNetNew找不到对应模型

File "/home/aistudio/PaddleOCR-release-2.7/ppocr/modeling/backbones/init.py", line 77, in build_backbone
assert module_name in support_dict, Exception(
AssertionError: when model typs is det, backbone only support ['MobileNetV3', 'ResNet', 'ResNet_vd', 'ResNet_SAST', 'PPLCNet', 'PPLCNetV3', 'PPHGNet_small']

更改两个studen的backbone参数

	Backbone:
        name: PPLCNetV3 # 将PPLCNetNew改为PPLCNetV3
        det: True # 添加det: True

训练时模型评估时，精度(precision)、召回率(recall)、F1值(hmean)都为0

[2023/03/24 11:24:16] ppocr INFO: best metric, hmean: 0, is_float16: False, precision: 0.0, recall: 0.0, fps: 6.358296325689887, best_epoch: 45

将config文件中的lr调高我设置的是0.02

ValueError: too many values to unpack (expected 4)

  File "/workspace/PaddleOCR/ppocr/postprocess/db_postprocess.py", line 229, in __call__
    src_h, src_w, ratio_h, ratio_w = shape_list[batch_index]
ValueError: too many values to unpack (expected 4)

将config文件中的cal_metric_during_train改为false