search

ortegatron / hand_detector_train

Hand Keypoint Detector trained with Tensorflow
36 stars 11 forks source link

Why are the hand keypoints used from synth hands dataset in the form of a 22x3 array? #3

Open shalini-maiti opened 4 years ago

shalini-maiti commented 4 years ago

From what I know, the open pose network outputs 22x2, i.e 2d coordinates for each keypoint. Why does it use 3d coordinates from the hand keypoints from the synth hands?

Jane333 commented 4 years ago

In the SynthHands dataset, the 3D positions are given in the depth camera space. For how to compute the 2D joint positions from these provided 3D positions, see the SynthHands FAQ here: https://handtracker.mpi-inf.mpg.de/projects/OccludedHands/SynthHands.htm