duke2market results

[JasAva]

Thanks for the inspiring implementation. :)

Have some trouble to reproduce the duke2market results using 2 GPUs, as I've achieved blew results using default parameters and pre-trained model: 1 GPU 47.8 / 71.4 (mAP/Rank-1) 2 GPUs 49.6 / 72.1 I've noticed a similar issue in closed ones but seems that 2 GPUs should be able to achieve the reported performances. My env is PyTorch 1.0, python 3.6 (made little changes to adjust to the Env, such as load pre-trained model, however, this should not affect the performances, as the pre-trained model is tested with reported performances).

Can you shed some insights on what could cause this problem? Thanks!

[flychen321]

I encountered the same problem, did you finally solve it?

[lsongx]

Sorry, I don't know how to solve this. For my experience, the BN layer affects the results a lot. So using 2 GPU makes a big difference since the BN becomes unsync-BN.