google-ai-edge / mediapipe

Cross-platform, customizable ML solutions for live and streaming media.
https://ai.google.dev/edge/mediapipe
Apache License 2.0
27.5k stars 5.15k forks source link

Issues with reliability / accuracy in dark rooms (low-light) #3208

Closed jsegeren closed 2 years ago

jsegeren commented 2 years ago

Issues with reliability / accuracy in dark rooms (low-light) Not surprisingly, we see that hand detection + tracking performance (both reliability - does it detect at all - and accuracy - does it track correctly) degrades significantly in a dark room / low lighting. (We also see some similar but opposite cases, i.e. bright lighting and light skin. i.e. Adverse environment lighting in general.)

We want to explore techniques that might help "squeeze out" better perf in these conditions. E.g. Image pre-processing - histogram equalization, or perhaps something else. Do you have any specific recommendations/guidance for us? Or perhaps there is more planned effort on your side to improve the out-of-box performance here?

Thank you for your help!

sureshdagooglecom commented 2 years ago

Hi @jsegeren , Could you share any reference videos or images w.r.t above issue.

google-ml-butler[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

jsegeren commented 2 years ago

Hi we are collecting some frames that we can share with you. That being said, this is quite easy to reproduce, even with your Codepen browser demo of MP Hands. E.g. move your hand quite quickly, even in decent lighting, if there is blur, the palm detection / keypoint estimation will struggle. Certainly a model issue (and of course a difficult problem to solve).

sureshdagooglecom commented 2 years ago

Hi @jsegeren , Could you please share any reference videos as discussed above.

google-ml-butler[bot] commented 2 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

google-ml-butler[bot] commented 2 years ago

Closing as stale. Please reopen if you'd like to work on this further.