search

VinAIResearch / CPM

💄 Lipstick ain't enough: Beyond Color-Matching for In-the-Wild Makeup Transfer (CVPR 2021)
https://thaoshibe.github.io/CPM
Apache License 2.0
367 stars 58 forks source link

The performance of the code can not match the picture in README #33

Closed tsuJohc closed 10 months ago

tsuJohc commented 10 months ago

Hi, Thanks for sharing the excellent work!

However, after successfully running the code, the results are visually worse than the demo picture in README. Are the bad results related to the code environment? Or the results are expected?

Demo picture in README:

result

Results(color+pattern) after I running:

c+p3

Results(color) after I running: ColorOnly

Results(pattern) after I running: patternonly

tsuJohc commented 10 months ago

After running the same code several times, I get a new result that is closer to the demo picture. Is there the randomness in the output of code? New results after I running: color+pattern2

thaoshibe commented 10 months ago

Hello, I think for Color-Only and Pattern-Only results seems OK. (Color-Only= Only transfer the color-makeup (e.g., lipstick); Pattern-Only= Only transfer the pattern (e.g., Moon-shape on forehead)).

For the randomness, I think you might try to vary the alpha of blending function here. Also, I'd suggest to check segmentation-mask results [https://github.com/VinAIResearch/CPM/blob/b12e83f195f6c3d378b93bb62da2fcdeb52f1312/main.py#L45] to see if it's consistent each time. Personally, I never run into cases where the results are not consistent (like you're facing). Maybe change the alpha in blending function should help!

tsuJohc commented 10 months ago

How should I change the alpha, lower or higher?

thaoshibe commented 10 months ago

You can vary it from 0 to 1.