Tf-pose-estimation: Do you have a mobilenet hand model?

[sachinsdate]

Do you have a mobilenet hand model?

Thanks
Sachin

[hybridherbst]

Would like to add to that: would be great if you can port the face model over, and/or the face-only code.
https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/doc/standalone_face_or_hand_keypoint_detector.md

[princefr]

Any update on this ?

[atyshka]

@ildoonet Hand detection would be a great feature. If you don't have the time to implement yourself, some of us from the community could contribute. Could you provide some information on how you converted the caffe body models to tensorflow? It seems the method used for hand detection is cropping around the wrist points and feeding each image into a CNN that outputs 11 points which are then constructed to form the hand?

[elbekri]

@atyshka https://arvrjourney.com/human-pose-estimation-using-openpose-with-tensorflow-part-1-7dd4ca5c8027 this should help you to understand how to convert caffe models to tensorflow.

[atyshka]

@elbekri Personally I dislike the caffe models, the VGG architecture is just too slow. I’m wondering if we could get some pointers on creating a mobilenet variant for the hand detection.

[elbekri]

I couldn't disagree on that @atyshka it is really slow but efficient and accurate.
However there is a lot of references in the ildoonet's README that should help you, but if you have any specific question i'll be happy to do some research/experimentation with you.
Btw Microsoft published a nice dataset for foot keypoint https://cmu-perceptual-computing-lab.github.io/foot_keypoint_dataset/

[HHIR]

@elbekri @atyshka any update?

[vnt1537]

Do you plan to include hand model in this? @ildoonet

[aa12356jm]

expecting