Shual

训练达不到论文中的效果

4

training from scratch（ic 15）80.11% training from MLT pretrained model（ic 15）84.7%左右，比论文和此repo的结果都低一个点。 实验 环境：2080Ti *2 pretrain on MLT， batchsize=10，n_epoch=320，schedule=[100,200], init_LR=1e-3。 预训练模型收敛时在ic15数据集上做测试可得hmean=82~83% finetune on ic15, batchsize=10， n_epoch=400, schedule=[200], init_LR=1e-4

sudo apt-get install python-dev # for python2.x installs sudo apt-get install python3-dev # for python3.x installs

when i run python script.py -g=... -s=... , I got two error. As follow: 1) Error: ('polygon has intersection sides', 2) Line in sample not valid. Sample: 1269 Line: I...

训练结果差

1

ICDAR2015数据集，训了50个epoch以后，eval结果任然是0.05左右。 而且分类loss和刚开始训的时候一样一直是0.99左右（如果我没记错的话）。 想请问一下，问题出在dataloader上还是损失函数那个部分？ 感谢！

Some files may be missing?

3

Thank you very much for open-sourcing this great work. Some files (such as /data , DataSet, config.py ...) may be missing here. If you have spare time, I hope you...

Wait for you, too. It's an amazing work.

RuntimeError: cannot unsqueeze empty tensor

3

`Traceback (most recent call last): File "eval_all.py", line 380, in main() File "eval_all.py", line 347, in main out, seg_pred, seg_map =net(img) File "/opt/conda/lib/python3.6/site-packages/torch/nn/modules/module.py", line 477, in __call__ result = self.forward(*input,...

AttributeError: Method AffineChannel is not a registered operator.

1

`WARNING cnn.py: 40: [====DEPRECATE WARNING====]: you are creating an object from CNNModelHelper class which will be deprecated soon. Please use ModelHelper object with brew module. For more information, please refer...