yxgeee
commented
4 years ago Hi,
No extra tricks, just randomness. I have re-trained it several times and the performances were all above 73%, so I used 73.1% in the final version. The only difference between SpCL+ (OpenUnReID) and SpCL in this repo is that SpCL+ adopts GeM pooling while SpCL adopts average pooling.
Best
Haijian Wang notifications@github.com于2020年10月20日 周二17:39写道：
Hi, I noticed that the USL performances on Market1501 dataset of your original arxiv version (mAP:72.6%) are slightly lower than that of your NeurIPS-2020 camera-ready version (mAP:73.1%), have you used some extral tricks? And what contributes to the amazing performance improvements of the OpenUnReID SpCL+ version compared to your original SpCL version? Thank you!
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/yxgeee/SpCL/issues/14, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACHSH2AM3Y2QYJ6TMWVPF33SLVLFTANCNFSM4SX6ZIMA .
Hi, I noticed that the USL performances on Market1501 dataset of your original arxiv version (mAP:72.6%) are slightly lower than that of your NeurIPS-2020 camera-ready version (mAP:73.1%), have you used some extral tricks? And what contributes to the amazing performance improvements of the OpenUnReID SpCL+ version compared to your original SpCL version? Thank you!