Table 2.

Pointing game evaluation measure with two ground-truth (GT) human annotations and recall across the different methods (Fig. 5).

GT Human3	Prec.	Recall	GT Human1	Prec.	Recall
Human1	0.81	0.77	Human 2	0.86	0.83
RISE [10]	0.68	0.48	RISE	0.78	0.52
Grad-CAM [9]	0.33	0.28	Grad-CAM	0.31	0.30
Occlusion [13]	0.29	0.31	Occlusion	0.30	0.38
LIME [8]	0.30	0.47	LIME	0.30	0.48