Table 4:

An ablation study on the benefit of MulT’s crossmodal transformers using CMU-MOSEI.).

Description	(Unaligned) CMU-MOSEI Sentiment
	${\text{Acc}}_7^h$	${\text{Acc}}_2^h$	F1h	MAEl	Corrh
Unimodal Transformers
Language only	46.5	77.4	78.2	0.653	0.631
Audio only	41.4	65.6	68.8	0.764	0.310
Vision only	43.5	66.4	69.3	0.759	0.343
Late Fusion by using Multiple Unimodal Transformers
LF-Transformer	47.9	78.6	78.5	0.636	0.658
Temporally Concatenated Early Fusion Transformer
EF-Transformer	47.8	78.9	78.8	0.648	0.647
Multimodal Transfomers
Only [V,A → L] (ours)	50.5	80.1	80.4	0.605	0.670
Only [L,A → V ] (ours)	48.2	79.7	80.2	0.611	0.651
Only [L,V → A] (ours) MulT mixing intermediate-	47.5	79.2	79.7	0.620	0.648
level features (ours)	50.3	80.5	80.6	0.602	0.674
MulT (ours)	50.7	81.6	81.6	0.591	0.691