mingminzhen
mingminzhen
I see. just need to do the following #if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600 #else #endif int file /include/caffe/common.cuh
@speedinghzl So do you use ohem threshold 0.6 or 0.7 to achieve the same performance?
@speedinghzl Actually, when I use my network in the new version, I can just get lower performance. I am not sure what happed? It seems you use DistributedDataParallel from apex,...
I use the original config given in fact.
thanks