assafshocher / ZSSR

"Zero-Shot" Super-Resolution using Deep Internal Learning
Other
401 stars 102 forks source link

About the image of paper #14

Open lkyee opened 5 years ago

lkyee commented 5 years ago

Hi, Thanks for your great work! I want to know how can i get the image like image I mean the image contains red and green parts. Look for your kind reply! Thank you!

assafshocher commented 5 years ago

If I remember correctly:

  1. We used a grayscale version of the image for all three channels.
  2. Applied ZSSR and EDSR to the same image.
  3. Then created a grascale map = (ZSSR-GT)^2 - (EDSR-GT)^2 where GT is the Ground Truth image. This means that positive values indicate pixels where ZSSR wins
  4. Normalized this map by it's max
  5. Used some spatial smoothing (maybe gaussian blur)
  6. Then take map (map>0) and add to the green channel, and map (map<0) to the red channel. (dont forget to clip to 0..1).