[cheeky feature request] detect funky masks (next-gen)

#23 · open · 0 comments

View on GitHub ↗

ghost

some are more easy (non human faces) <img src="https://user-images.githubusercontent.com/415238/158080922-efb00511-195a-41ff-8978-0a16618ec6e5.jpg" /><br/> some are human based but distorted or have an obvious (non-mask) overlay <img src="https://user-images.githubusercontent.com/415238/158080983-95ab2c19-3d94-40f7-a4a9-14f46ea2f285.jpg" /><br/> this is probably still detectable <img src="https://user-images.githubusercontent.com/415238/158081061-1e96189c-708d-4214-a070-925bdd652f58.jpg" /><br/> challenge accepted... (partial mask is part of the graphics) ;) <img src="https://user-images.githubusercontent.com/415238/158081075-5d8db509-1c71-4776-a328-7f396a613663.jpg" /><br/> source: https://www.ebay.com/sch/i.html?_nkw=funny+face+mask seams and joints (requires high resolution), nose shadow non-consistent with other shadow cast on face, and skin tone difference would normally improve (heuristic) results. I wonder how ML would handle it..

Comments