search

WenmuZhou / PSENet.pytorch

A pytorch re-implementation of PSENet: Shape Robust Text Detection with Progressive Scale Expansion Network
GNU General Public License v3.0
462 stars 138 forks source link

Traning on tianchi data #2

Closed NormalAppler closed 5 years ago

NormalAppler commented 5 years ago

Hi,When I use the code to train TianChi Data,It doesn't work.Many errors happened.So have you trained on TianChi Data?Which code should I modify to train TianChi data.Could you give me some advice ? Thank you

WenmuZhou commented 5 years ago

能给个链接吗，天池的数据

NormalAppler commented 5 years ago

链接：https://pan.baidu.com/s/11kM4yC_qOrQabNjM6d04rQ 提取码：ue1p 这是天池数据 9000张训练，1000张测试

WenmuZhou commented 5 years ago

你需要改一下这里 https://github.com/WenmuZhou/PSENet.pytorch/blob/6c435d95f71c6022ba1a97ef1c9b371b8b960132/dataset/data_utils.py#L164-L165

NormalAppler commented 5 years ago

你需要改一下这里

PSENet.pytorch/dataset/data_utils.py

Lines 164 to 165 in 6c435d9

d = pathlib.Path(x) label_path = os.path.join(datadir, 'gt', ('gt' + str(d.stem) + '.txt'))

这个地方改了，而且还改了那个判断数据格式正则表达式的地方，依然报错，训练的时候没事，就是在进行对测试集进行验证的时候，进度条结束以后就会报错

WenmuZhou commented 5 years ago

你验证的gt换了吗，我用的验证的脚本是icdar2015的，你可能要根据天池的做一下修改

NormalAppler commented 5 years ago

我再查看一下，谢谢

NormalAppler commented 5 years ago

作者，你好，发现天池的数据集txt文件中的坐标是逆时针的，而IDCAR的数据是顺时针的，请问再训练的时候有影响吗，需要换成顺时针的吗

WenmuZhou commented 5 years ago

没有试过这样的数据，不可以训练一下，看会不会报错

NormalAppler commented 5 years ago

好的，谢谢