Figure 6
From: Three-dimensional atrous inception module for crowd behavior classification

Confusion matrix on the Crowd-11 dataset using X3D with varying loss functions. (a) RGB with SoftMax. (b) RGB with binary CE. (c) RGB with separation loss. (d) OF with SoftMax. (e) OF with binary CE. (f) OF with separation loss. (g) Two-stream with SoftMax. (h) Two-stream with binary CE. (i) Two-stream with separation loss.