I want to achieve the U-net in pytorch,but when the batch_size>1,it point out that"out of menory",but when i achieve in tensorflow ,the batch_size could be 15,in the same structure.?i use GTX1080ti with 10GB menory
It’s hard to tell any differences between your PyTorch and TF code without any code samples.
Did you implement both solutions on your own or did you use some from e.g. github?
In the latter case did you compare the network structure, i.e. number of kernels, kernel size, etc.?