trian

[sakurariven]

trian

[wangyuxin87]

Waiting for your response!

[SakuraRiven]

@wangyuxin87 hi, the code has been only tested on my environment, which is shown in README. So if you are using another environment which causes that bug, change num_workers to 0 and try again. I hope this helps : )

[wangyuxin87]

Thanks for your code and response!
I have some new issue, can you help me?

root@bfd0e7a2dba1:/home/yxwang/pytorch/EAST# CUDA_VISIBLE_DEVICES=6,7 python3 train.py
Traceback (most recent call last):
File "train.py", line 66, in
train(train_img_path, train_gt_path, pths_path, batch_size, lr, num_workers, epoch_iter, save_interval)
File "train.py", line 35, in train
for i, (img, gt_score, gt_geo, ignored_map) in enumerate(train_loader):
File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 615, in next
batch = self.collate_fn([self.dataset[i] for i in indices])
File "/usr/local/lib/python3.5/dist-packages/torch/utils/data/dataloader.py", line 615, in
batch = self.collate_fn([self.dataset[i] for i in indices])
File "/home/yxwang/pytorch/EAST/dataset.py", line 384, in getitem
lines = f.readlines()
File "/usr/lib/python3.5/encodings/ascii.py", line 26, in decode
return codecs.ascii_decode(input, self.errors)[0]
UnicodeDecodeError: 'ascii' codec can't decode byte 0xef in position 0: ordinal not in range(128)

Waiting for your response!

[SakuraRiven]

Putting #coding=utf-8 in the 1st line may help

[wangyuxin87]

I try # coding=utf-8 or # coding:utf-8 in the 1st line of dataset.py, it does not work.... Do you have any other idea? Thanks

[SakuraRiven]

you may refer to https://stackoverflow.com/questions/35028683/python3-unicodedecodeerror-with-readlines-method

[wangyuxin87]

Thanks! It has worked. Do you add ICDAR 2013 in your training images? I think it will improve the performance.

[SakuraRiven]

The data distribution in 13 and 15 still exists difference, so you can try and welcome to share the results : )

[wangyuxin87]

The results get worse when I change the basic network to Resnet50, do you have any idea?
By the way, it obtains the same performance with extra dataset(ICDAR2013).