out of memory about two-stream-pytorch

[bryanyzhu]

Hi, thank you for your interest.

(1) For the OOM issue, maybe reduce the batch size a little bit and see. If it still comes up, then maybe there is memory leakage problem. We can investigate more on the problem.

(2) For multi-GPU, I think it is straightforward to extend. Just make

model = torch.nn.DataParallel(model).cuda()

And it should work. But pay attention to batch size, the total batch size should be less than 256, otherwise maybe there is diverge problem.