Hi, thanks for your nice work.
On a large dataset like ImageNet, proposed self-attention mechanism consumes large amounts of memory because of unfold operation (im2col).
One would want to share point-wise feature vector among sliding windows.
Do you have any idea?
Thanks for comments.
I didn't think of a more efficient way. I will think about the effective self-attention mechanism. If you have a good way, send pull request.
Hi, thanks for your nice work. On a large dataset like ImageNet, proposed self-attention mechanism consumes large amounts of memory because of unfold operation (im2col). One would want to share point-wise feature vector among sliding windows. Do you have any idea?