DeepGuidedFilter
DeepGuidedFilter copied to clipboard
wuhuikai
What is low resolution image in context of semantic segmentation
In semantic segmentation the default setup is:
- input image
- backbone which returns features at strides 2, 4, 8, 16, 32
- decoder which combines features and produces logits with stride-2 (or 4 as in DeepLabV3+)
- head which resizes logits to image size and makes predictions (conv+softmax/sigmoid)
So what should be a low resolution image for ConvGuidedFilter in this case? Should i resize original image to logits size (stride 2 or 4) or i should use stride-2/4 features from backbone?
Guidance image: img with HxW Low-res input: img with H/4xW/4 Low-res output: logits with H/4 xW/4
On Mar 25, 2022, at 7:33 PM, Shkarupa Alex @.***> wrote:
In semantic segmentation the default setup is:
input image backbone which returns features at strides 2, 4, 8, 16, 32 decoder which combines features and produces logits with stride-2 (or 4 as in DeepLabV3+) head which resizes logits to image size and makes predictions (conv+softmax/sigmoid) So what should be a low resolution image for ConvGuidedFilter in this case? Should i resize original image to logits size (stride 2 or 4) or i should use stride-2/4 features from backbone?
— Reply to this email directly, view it on GitHub https://github.com/wuhuikai/DeepGuidedFilter/issues/34, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABULPVNCQXZY77RNFQQSTW3VBWQBXANCNFSM5RT5HZDQ. You are receiving this because you are subscribed to this thread.
Low-res image obtained by just resizing original one? Did you test what interpolation (linear/cubic/area) works better?
Yes, I only tried bilinear.
On Mar 25, 2022, at 8:48 PM, Shkarupa Alex @.***> wrote:
Low-res image obtained by just resizing original one? Did you test what interpolation (linear/cubic/area) works better?
— Reply to this email directly, view it on GitHub https://github.com/wuhuikai/DeepGuidedFilter/issues/34#issuecomment-1078996784, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABULPVNKVCG54VGPEFFOUXDVBWY3PANCNFSM5RT5HZDQ. You are receiving this because you commented.