Paddle OCR 训练报错
发布于2021-09 浏览:1595 回复:2
0
收藏

你好;

执行

aistudio@jupyter-166070-2374013:~/PaddleOCR/PaddleOCR$ python tools/train.py -c configs/det/det_mv3_db.yml
[2021/09/15 17:35:06] root INFO: Architecture :
[2021/09/15 17:35:06] root INFO: Backbone :
[2021/09/15 17:35:06] root INFO: model_name : large
[2021/09/15 17:35:06] root INFO: name : MobileNetV3
[2021/09/15 17:35:06] root INFO: scale : 0.5

。。。。。。。。。

[2021/09/15 17:35:06] root INFO: loader :
[2021/09/15 17:35:06] root INFO: batch_size_per_card : 16
[2021/09/15 17:35:06] root INFO: drop_last : False
[2021/09/15 17:35:06] root INFO: num_workers : 8
[2021/09/15 17:35:06] root INFO: shuffle : True
[2021/09/15 17:35:06] root INFO: use_shared_memory : False
[2021/09/15 17:35:06] root INFO: train with paddle 2.1.2 and device CPUPlace
[2021/09/15 17:35:06] root INFO: Initialize indexs of datasets:['./train_data/icdar2015/text_localization/train_icdar2015_label.txt']
[2021/09/15 17:35:06] root INFO: Initialize indexs of datasets:['./train_data/icdar2015/text_localization/test_icdar2015_label.txt']
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.conv.conv.weight [8, 3, 3, 3] not matched with loaded params conv5_linear_bn_variance (24,) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.conv.bn.bias [8] not matched with loaded params conv6_se_1_weights (16, 64, 1, 1) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.conv.bn._mean [8] not matched with loaded params conv12_expand_bn_offset (240,) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.conv.bn._variance [8] not matched with loaded params conv6_expand_bn_scale (64,) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.stage0.0.expand_conv.conv.weight [8, 8, 1, 1] not matchellllld with loaded params conv8_linear_bn_variance (40,) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.stage0.0.expand_conv.bn.bias [8] not matched with loaded params conv10_expand_weights (96, 40, 1, 1) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.stage0.0.expand_conv.bn._mean [8] not matched with loaded params conv10_linear_bn_variance (40,) !
[2021/09/15 17:35:06] root INFO: The shape of model params backbone.stage0.0.expand_conv.bn._variance [8] not matched with loaded params conv16_expand_bn_scale (480,) !

 

这个报错会影响训练结果吗

收藏
点赞
0
个赞
共2条回复 最后由用户已被禁言回复于2022-04
#3草原狼legend回复于2021-09

https://github.com/PaddlePaddle/PaddleOCR/issues/4105

还是需要加载预训练模型的,这个问题提到issue里面了,而且我看是CPU训练的,资源应该不太够

0
#2回复于2021-09

没有发帖,我自己回答一下把,首先这个问题是加载预训练模型导致, 官网给出了错误的下载地址,导致加载模型和训练的网络结构有冲突。加载预训练模型是为了迁移学习,不加载好像也可以,只是需要很长的训练时间才能提高准确率。

0
TOP
切换版块