AlexHex7/Non-local_pytorch

3D pooling

protein27 opened this issue 6 years ago · 2 comments

protein27 commented 6 years ago

Non-local_pytorch/lib/non_local_simple_version.py

Line 59 in 589dde8

self.g = nn.Sequential(self.g, max_pool(kernel_size=2))

In the paper, the pooling is only in the spatial domain. I think the kernel size is not set correctly when the input contains temporal dimension.

https://github.com/facebookresearch/video-nonlocal-net/blob/b273c446e8e10dbaec266520e4005d27d7052125/lib/models/nonlocal_helper.py#L40

AlexHex7 commented 6 years ago

@protein27 Thanks for pointing it out! I will correct it when I have time.

AlexHex7 commented 6 years ago

I have corrected the code .