Hierarchical Attention-Based Multimodal Fusion Network for Video Emotion Recognition

Table 5

Accuracy of emotion recognition of different modals on Ekman and VideoEmotion-8 datasets.

Ekman			VideoEmotion-8
Convolution layers	Event (%)	Object (%)	Scene (%)	Event (%)	Object (%)	Scene (%)
No attention	42.45	36.43	40.95	48.10	46.45	46.33
L1	44.14	41.42	44.41	51.34	49.88	49.14
L2	45.78	41.14	44.69	53.18	49.63	49.39
L3	45.23	40.33	43.60	52.81	48.90	49.02