Multiple Context Learning Networks for Visual Question Answering

Table 2

The results of ablating the context learning modules on VQA v2.0 and GQA validation sets.

Model	Module	GQA	VQA v2.0
		All	All	Y/N	Num	Other
1	Without all	53.08	54.60	69.79	36.02	47.50
2	Only VCL	53.45	55.13	69.82	36.09	47.99
3	Only TCL	53.50	55.53	69.82	36.44	49.72
4	Only VTCL	58.63	62.07	79.79	42.67	53.73
5	TCL + VTCL	63.88	65.17	82.88	44.68	57.15
6	VCL + VTCL	59.04	62.72	79.33	43.29	55.23
7	VCL + TCL	53.72	55.76	71.01	36.37	49.29
8	Full modules	64.48	65.68	83.40	45.57	57.53