train with swin backbone

[shengyi4]

I am tring to reproduce the result for uppernet + swin transformer. However, when I run: python tools/train.py configs/swin/upernet_swin_base_patch4_window12_512x512_160k_ade20k_pretrain_384x384_22K.py . It shows an error like this: Traceback (most recent call last): File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\verifier.py", line 43, in VerifyCode compile(textwrap.dedent(code).encode('UTF-8'), '', 'exec') File "", line 2 work_dir = './work_dirs\upernet_swin_base_patch4_window12_512x512_160k_ade20k_pretrain_384x384_22K' ^ SyntaxError: (unicode error) 'unicodeescape' codec can't decode bytes in position 11-12: truncated \uXXXX escape

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\verifier.py", line 46, in VerifyCode ast.parse(textwrap.dedent(code.lstrip('\n')).lstrip(), '', 'exec') File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\ast.py", line 50, in parse return compile(source, filename, mode, flags, File "", line 1 work_dir = './work_dirs\upernet_swin_base_patch4_window12_512x512_160k_ade20k_pretrain_384x384_22K' ^ SyntaxError: (unicode error) 'unicodeescape' codec can't decode bytes in position 11-12: truncated \uXXXX escape

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\verifier.py", line 50, in VerifyCode compile(normalized_code.encode('UTF-8'), '', 'exec') File "", line 1 work_dir = './work_dirs\upernet_swin_base_patch4_window12_512x512_160k_ade20k_pretrain_384x384_22K' ^ SyntaxError: (unicode error) 'unicodeescape' codec can't decode bytes in position 11-12: truncated \uXXXX escape

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "C:\Users\sheng\Desktop\mmsegmentation\tools\train.py", line 209, in main() File "C:\Users\sheng\Desktop\mmsegmentation\tools\train.py", line 135, in main cfg.dump(osp.join(cfg.work_dir, osp.basename(args.config))) File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\mmcv\utils\config.py", line 564, in dump f.write(self.pretty_text) File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\mmcv\utils\config.py", line 502, in prettytext text, = FormatCode(text, style_config=yapf_style, verify=True) File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\yapf_api.py", line 188, in FormatCode reformatted_source = FormatTree( File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\yapf_api.py", line 151, in FormatTree return reformatter.Reformat(_SplitSemicolons(llines), verify, lines) File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\reformatter.py", line 105, in Reformat return _FormatFinalLines(final_lines, verify) File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\reformatter.py", line 413, in _FormatFinalLines verifier.VerifyCode(formatted_code[-1]) File "C:\Users\sheng\miniconda3\envs\open-mmlab\lib\site-packages\yapf\yapflib\verifier.py", line 52, in VerifyCode raise InternalError(sys.exc_info()[1]) yapf.yapflib.verifier.InternalError: (unicode error) 'unicodeescape' codec can't decode bytes in position 11-12: truncated \uXXXX escape (, line 1)

It worked fine with other configures. Does anyone else encounter this problem? Thanks.

[MengzhangLI]

work_dir = './work_dirs\upernet_swin_base_patch4_window12_512x512_160k_ade20k_pretrain_384x384_22K'

I think it is caused by wrong path, perhaps caused by different OS environment (windows and linux)?

[shengyi4]

Thanks for your feedback! I am using windows. I didn't specify the path in the configuration file as well. It worked for other configurations (deep lab v3, segformer, and so on). But it doesn't work for swin.

[MengzhangLI]

OK, just set correct path and it could work normally in windows too.

[shengyi4]

Thanks! It works when I set the path in the command line.

[Donfugui]

excuse me ,the same error happens to me now, i wonder how you solve the issue exactly, thanks

[scm-later]

the same error happens to me now

[duduzai2019]

我现在也发生了同样的错误

Hi, excuse me ,the same error happens to me now, i wonder how you solve the issue exactly, thanks,

[duduzai2019]

excuse me ,the same error happens to me now, i wonder how you solve the issue exactly, thanks

excuse me ,the same error happens to me now, i wonder how you solve the issue exactly, thanks

[duduzai2019]

the error is unicoder, so i find the dataset corresponds files, replace the data_root to \ , after can train nomally

[zznew]

因为Swin Transformer的backone为upernet, os.path.join()会使用"\"来拼接路径，这样就会出现\u，而\uXXXX代表一个unicode编码。所以找到代码中的os.path.join()全部修改就好了