DeepPrior++

[mjsobrep]

https://github.com/moberweger/deep-prior-pp provides depth based detections for hands which look promising. It should be possible to make it work for images the way we get them.

It might be a good idea to use updated libraries: moberweger/deep-prior-pp#23

If you start here: https://github.com/moberweger/deep-prior-pp/blob/11f585f73d2c7957c95db302b770a3b4962c0386/src/test_realtimepipeline.py#L74 and look at how the filedevice is implemented: https://github.com/moberweger/deep-prior-pp/blob/11f585f73d2c7957c95db302b770a3b4962c0386/src/util/cameradevice.py#L348 then you can see pretty easily how one could tie into our hdf5 architecture. This ends up getting processed here: https://github.com/moberweger/deep-prior-pp/blob/11f585f73d2c7957c95db302b770a3b4962c0386/src/util/realtimehandposepipeline.py#L221

You will need to detect the hand in the frame : https://github.com/moberweger/deep-prior-pp/blob/11f585f73d2c7957c95db302b770a3b4962c0386/src/util/realtimehandposepipeline.py#L164
Then you will need to crop it: https://github.com/moberweger/deep-prior-pp/blob/11f585f73d2c7957c95db302b770a3b4962c0386/src/util/realtimehandposepipeline.py#L168
then you will need to estimate pose: https://github.com/moberweger/deep-prior-pp/blob/11f585f73d2c7957c95db302b770a3b4962c0386/src/util/realtimehandposepipeline.py#L197