RuntimeError: Expected object of backend CPU but got backend CUDA for argument #2 'weight'

vision
1

I ran into this problem while running my network. Do you guys have any thought? here are the code
##network##

image
image997×853 49.3 KB

##training##
image
image969×611 31.6 KB

##error##
image
image837×542 32.1 KB

Thanks alot!!

2

If you are pushing tensors to the device, you have to reassign them:

clips = clips.to(device)
optical_flows = optical_flows.to(device)
...

Also, it’s recommended to call the model directly (net(clips, optical_flows)) instead of the .forward() method. This is unrelated to your current issue, but could avoid other problems in the future. :wink:

9 Likes
3

i solved my problem
thank you!