Region Dual Attention-Based Video Emotion Recognition

Table 6

Top-1 accuracy (%) comparing with state-of-the-art methods on Ekman-6 and VideoEmotion-8.

Method	Ekman (%)	VideoEmotion-8 (%)
Emotion in context [12]	51.8	50.6
Xu et al. [30]	50.4	46.7
Kernelized feature [34]	54.4	49.7
Concept selection [35]	54.40	50.82
Graph-based network [36]	55.01	51.77
Ours	58.07	54.22

We also evaluate the performance of our method on the SFEW data set for cross-validation purposes.