search

zhang-can / ECO-pytorch

PyTorch implementation for "ECO: Efficient Convolutional Network for Online Video Understanding", ECCV 2018
BSD 2-Clause "Simplified" License
188 stars 93 forks source link

about the inference demo #23

Closed Usernamezhx closed 6 years ago

Usernamezhx commented 6 years ago

sooorry to bother you again :p I want to test the online recognition. but the inference erro like this: **RuntimeError: cuda runtime error (2) : out of memory at /pytorch/aten/src/THC/generic/THCStorage.cu:58** only i set the batchsize to 2 and then it can work. but I find the eval will success when the batchsize is 15. below is the code: 1 2 I test it on UCF101. thanks in advande

Usernamezhx commented 6 years ago

input_var = data.cuda() Does it matter? I find your code the input data run on the cpu. but when i change it to input_var = data it will show me that : _20180928134611

Usernamezhx commented 6 years ago

I think I find the problem. I am not familar to pytorch. I only have one graph card . so I replace the code model = torch.nn.DataParallel(model, device_ids=args.gpus).cuda() to mode.cuda() . when I recover it was normal........

forbiddenname commented 4 years ago

Hi, how to test the online recognition? Is Real-time Video Streaming ？