Fig. 5
From: ICANet: a simple cascade linear convolution network for face recognition
Impact of the histogram block size on ICANet: Assuming that the length and width of the chunks were equal in the ICANet block histogram stage, and both used overlapping and non-overlapping methods, the block size gradually increased from 8 ∼20. The experimental results show that the histogram block size had an impact on ICANet, and the face recognition system had the highest performance when the block size was 16×16