CVPR 2017

[arxiv 1704.02612] BigHand2.2M Benchmark: Hand Pose Dataset and State of the Art Analysis [PDF] [notes]

Shanxin Yuan, Qi Ye, Bjorn Stenger, Siddhand Jain, Tae-Kyun Kim

read 12/05/2017

Synthesis

6 6D Magnetic sensors used to generate 2.2Million annotated **depth frames ** used to generate the BigHand 2.2M benchmark dataset

290K egocentric frames

3d model used

Results

15- 20mm average error on existing benchmarks

Remarks

magnetic sensors already mentionned in other paper as cheap efficient, and surprisingly not widely used.

Considers state of the art other paper from Tae-Kyun Kim