search

csxmli2016 / DFDNet

Blind Face Restoration via Deep Multi-scale Component Dictionaries (ECCV 2020)
914 stars 216 forks source link

How to let crop and landmark detection run on gpu while running restoration on cpu #47

Closed Michaelwhite34 closed 3 years ago

Michaelwhite34 commented 3 years ago

I have the question because my gpu doesn't have enough memory.

Michaelwhite34 commented 3 years ago

Another question,is it possible to set some parameters manually ? For some highly compressed videos,there will be some kind of noise in the face (which can't be eliminated by denoise method),DFDNet will take it as face feature and produce some kind of freckle in the face.So can we manually set the face to be smooth (without much speckle )？

csxmli2016 commented 3 years ago

I have the question because my gpu doesn't have enough memory.

Hi, you can run the first two steps with gpu. Then with the detected and cropped results saved in the filepath, you can comment the two steps and directly run the restoration process with cpu.

csxmli2016 commented 3 years ago

Another question,is it possible to set some parameters manually ? For some highly compressed videos,there will be some kind of noise in the face (which can't be eliminated by denoise method),DFDNet will take it as face feature and produce some kind of freckle in the face.So can we manually set the face to be smooth (without much speckle )？

Good suggestion. Maybe it is better to consider this degradation in conducting training pairs.