xmengli/H-DenseUNet

multi GPU problem

JohnleeHIT opened this issue · 0 comments

Hello, I meet some problem when training your code, I debuged the code and find the problem, it attribute to the function "make_parallel", once i skip the process and use just one gpu, no error. However, when add this item, I get :
"could not set cudnn tensor descriptor: CUDNN_STATUS_BAD_PARAM"
Any idea why this happens? Thanks.