Multiscale Convolutional Neural Networks for Hand Detection

Table 3

Average Precision (AP) on VIVA L1 and L2 dataset and comparison with previous methods.

Method	L1 set	L2 set
CNNRegionSampling [9]	66.8%	57.8%
ACF Depth4 [6]	70.1%	60.1%
YOLO [10]	76.4%	69.5%
FRCNN [11]	90.7%	86.5%
Our model (Multiscale Fast R-CNN)	92.8%	84.7%