文章在本文中尝试去回答两个问题：
1、What is the current state of 3D hand pose estimation?
2、And, what are the next challenges that need to be tackled?
结合了2017年的竞赛，Hand In the Million Challenge(HIM2017)
three tasks : single frame, tracking, during object interaction
得到了以下结论：
(1) isolated 3D hand pose estimation achieves low mean errors (10 mm) in the view point range of [40, 150] degrees, but it is far from being solvedforextremeviewpoints;
(2) 3D volumetric representations outperform 2D CNNs, better capturing the spatial structureofthedepthdata;
(3) Discriminative methods generalize poorly to unseen hand shapes
(4) While joint occlusions pose a challenge for most methods, explicit modeling of structure constraints can significantly narrow the gap between errors on visible and occluded joints.
cross-benchmark testing is poor : view point, hand shape, self-occlusion, occlusion caused by objects.
network architectures, preprocessing strategies, data representations