Can someone explain how to use spatial_scale in the torchvision.ops.roi_pool( *input* , *boxes* , *output_size* , *spatial_scale=1.0* )) using some example. Let’s say my original image is 800*800 and subsampling ratio is 16, so feature map size is 50*50. Now in the input to torchvision.ops.roi_pool we give feature map and boxes are in image coordinate scale.
So spatial_scale should be equal to subsampling ratio?