search

solarorb93 / BetaSuite

63 stars 6 forks source link

skin filtering #38

Open solarorb93 opened 2 years ago

solarorb93 commented 2 years ago

Add option for full-body skin filtering

solarorb93 commented 2 years ago

NudeNet doesn't support this, obviously. I tried using HSV/YCrCb thresholding, got really pretty poor results. Also used the WillBrennan/SemanticSegmentation pytorch net - results were better but still not really good enough for BetaSuite use. Some combination of several approaches maybe someday could work but for now it doesn't really seem doable.

hostilesponge commented 2 years ago

I found this which may be of interest. For something like this I think you would need to create and train a model to differentiate between covered and exposed skin. It would probably be really tricky to train it, because you'd have to account for different skin tones, bad/weird lighting, strange angles and probably other things I'm not thinking of.

solarorb93 commented 2 years ago

I tried a few already-trained skin detection models with very little success. Which is a shame, because this could be a very cool feature. If I could reliably detect person vs nonperson (even if I can't distinguish exposed vs covered), that's still something I could use for very hard censoring, but I'm not optimistic about being able to do that only with edge detection...

On Fri, Aug 19, 2022, 8:28 PM hostilesponge @.***> wrote:

I found this https://www.researchgate.net/publication/271822998_Free-Shape_Polygonal_Object_Localization which may be of interest. For something like this I think you would need to create and train a model to differentiate between covered and exposed skin. It would probably be really tricky to train it, because you'd have to account for different skin tones, bad/weird lighting, strange angles and probably other things I'm not thinking of.

— Reply to this email directly, view it on GitHub https://github.com/solarorb93/BetaSuite/issues/38#issuecomment-1221222644, or unsubscribe https://github.com/notifications/unsubscribe-auth/AZSKWAYBCM7JYPQW6BTQUXDV2BGE5ANCNFSM55S3NOOA . You are receiving this because you authored the thread.Message ID: @.***>