search

snowzach / doods2

API for detecting objects in images and video streams using Tensorflow
MIT License
230 stars 30 forks source link

[FR:] Facedetection models possible? #34

Closed ozett closed 1 year ago

ozett commented 2 years ago

fiddling with deepstack i came to the idea that doods2 could have face-models. i see deepstack make use of ARCface and MobileFaceNet

https://github.com/johnolafenwa/DeepStack/blob/5da0520b855b8db0692c639a1d4d8c83a49e9883/intelligencelayer/shared/recognition/networks.py#L77-L78

https://github.com/johnolafenwa/DeepStack/blob/5da0520b855b8db0692c639a1d4d8c83a49e9883/intelligencelayer/shared/recognition/networks.py#L224

but i dont see how this could be implemented in doods2

ozett commented 2 years ago

could that work? https://github.com/timesler/facenet-pytorch

ozett commented 2 years ago

or this? https://github.com/deepinsight/insightface?ref=https://githubhelp.com

ozett commented 2 years ago

i found deepstack in your code... https://github.com/snowzach/doods2/blob/9dcda621cec9c1ddb487737383194f0c02dcde99/detectors/deepstack.py#L13

maybe facedetection from deepstack could be made usable?

https://github.com/johnolafenwa/DeepStack/blob/5da0520b855b8db0692c639a1d4d8c83a49e9883/intelligencelayer/shared/face.py#L26

dbeckers99 commented 2 years ago

I'm on the same way. I think using a different tensorflow model should be easy to implement.

My main problem is: I want to train the existing tensorflow model with false positive person detection images to make the detection result more reliable. But training Tensorflow models isn't that easy, you have to learn a lot about.

Anyone has an easy walkthrough to share?

snowzach commented 1 year ago

I wish there was on.. sorry.. Gonna close this to clean up issues for now.