Closed wangchust closed 6 years ago
Yes, I have the same problem. I think better performance of 4 GPU comes from that the average process of derivative on different GPUs .
Please check this section on our website.
Thank you @bnulihaixia and @Cysu . This open-reid framework is really an exitcing work!!!
Does anyone encounter this phenomenon, results on 4 GPUs are better than on 2GPUs. Results on 2GPUs are better than on 1 GPU. The other parameters are all kept same, including learning rate. Does anyone know why?