Can the CUDA memory occupied be further reduced? #33
Description
I understand that the memory usage of refine net and score net should not be high? Is it mainly or does the rendering part consume a lot of video memory? Have you counted the memory usage of rendering? Is there a better way to render it?
If you have any insights, I'd be delighted to discuss them with you!