Cannot obtain the reported performance by directly running the run.sh

[Reply1999]

Hi, I would like to thank you for releasing the codes in the first place.

Following the readme.md, we directly run the run.sh without any modification, but fail to obtain the reported performance. Can you help us figure out what the problem is?

To be more specific, we obtain Mean AP: 54.3% and top1: 77.3% (best) after training for 30 epoches by running the run.sh for SSG in Duke->Market1501, which should have Mean AP: 58.3% and top1: 80.0%.

[yihongXU]

Hi,
Would you mind telling me which parameters did you use for getting Mean AP: 54.3% and top1: 77.3% (best) ? And which version of pytorch did you use? You used two gpus or just one?
thank you!

Sincerely,

[OasisYang]

The parameters are the same as described in the code, I use pytorch=0.4.1 with two gpus.

[yihongXU]

The parameters are the same as described in the code, I use pytorch=0.4.1 with two gpus.

Hi thanks for your reply. I'll go try pytorch=0.4.1, torchvision=0.2.0, and as the settings in the code, I will try batch_size=128 instead of batch_size=32 recommended in the readme, which gives out an error during training.

[X-funbean]

Hi, I tried Duke->Market with default SSG config on 2 RTX2080Ti and I got the following result:

Mean AP: 54.1%
CMC Scores  market1501
top-1          76.1%
top-5          86.9%
top-10         90.2%

 * Finished epoch  30  top1: 76.1%  best: 76.5%

It seems that this result matches with those in several issues, which is apparently lower than what the paper has reported. Has anyone actually obtained the reported performance?