Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Adding support for rank cython in Windows, fixing pytorch deprecation warnings #412

Merged
merged 4 commits into from
Feb 9, 2021

Conversation

lablabla
Copy link

@lablabla lablabla commented Feb 3, 2021

Following suggestion here
ansys/pymapdl#14 (comment)
I got the cython to run on Windows and checked on Linux as well

Tested on Windows 64bit & Ubuntu 18.04

@KaiyangZhou
Copy link
Owner

Thanks so much for fixing this long-standing issue for Windows users!

One more step to go: can you please test the performance of resnet50 (you can download the weights here https://kaiyangzhou.github.io/deep-person-reid/MODEL_ZOO.html) on Market1501/DukeMTMC-reID using both the evaluation codes before and after the changes? I'd like to see whether the calculation precision will be affected.

@lablabla
Copy link
Author

lablabla commented Feb 5, 2021

I might be doing something wrong but I can't load the pretrained weights
I'm running this

python scripts/main.py --config-file configs/im_r50_softmax_256x128_amsgrad.yaml model.load_weights /home/devel/Downloads/resnet50_fc512_market_xent.pth.tar test.evaluate True

I've downloaded resnet50_fc512_market_xent.pth.tar from https://mega.nz/file/EaZjhKyS#lBvD3vAJ4DOmElZkNa7gyPM1RE661GUd2v9kK84gSZE

and I'm getting

Traceback (most recent call last):
  File "/home/devel/dev/deep-person-reid/torchreid/utils/torchtools.py", line 84, in load_checkpoint
    checkpoint = torch.load(fpath, map_location=map_location)
  File "/home/devel/.conda/envs/torchreid/lib/python3.7/site-packages/torch/serialization.py", line 595, in load
    return _legacy_load(opened_file, map_location, pickle_module, **pickle_load_args)
  File "/home/devel/.conda/envs/torchreid/lib/python3.7/site-packages/torch/serialization.py", line 774, in _legacy_load
    result = unpickler.load()
UnicodeDecodeError: 'utf-8' codec can't decode byte 0x98 in position 0: invalid start byte

Getting the same error when using the im_r50fc512_softmax_256x128_amsgrad.yaml config file
This is happening in the both version, before and after the change. looking for solution, probably not running it correctly

@KaiyangZhou
Copy link
Owner

I also met the UnicodeDecodeError when using resnet50. Hmm, that's weird.

Maybe try osnet, which works for me

python scripts/main.py \
--config-file configs/im_osnet_x1_0_softmax_256x128_amsgrad_cosine.yaml \
--root PATH_TO_DATA \
model.load_weights saved-models/osnet/osnet_x1_0_market_256x128_amsgrad_ep150_stp60_lr0.0015_b64_fb10_softmax_labelsmooth_flip.pth \
test.evaluate True

@lablabla
Copy link
Author

lablabla commented Feb 9, 2021

This one worked. please see the attached logs. Only difference (aside from the version) is the Speed, 0.0634 sec/batch vs 0.0638 sec/batch, which I guess is negligible

1.3.4.log
1.3.5.log

This was run on Ubuntu 18.04

Windows 1.3.5 (Had to lower the batch size since it's a laptop with less memory on the GPU and kept getting pytorch out of memory issues)
win 1.3.5.log

@KaiyangZhou
Copy link
Owner

Great!

The speed isn't really an issue because you normally get different speeds in different runs.

I'll merge this code with the master branch.

Thanks!

@KaiyangZhou KaiyangZhou closed this Feb 9, 2021
@KaiyangZhou KaiyangZhou reopened this Feb 9, 2021
@KaiyangZhou KaiyangZhou merged commit d064f4c into KaiyangZhou:master Feb 9, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants