EURASIP Journal on Advances in Signal Processing

Table 1 Different models experiment on RAF-DB dataset

From: Research on real-time teachers’ facial expression recognition based on YOLOv5 and attention mechanisms

Method	Model size	Parameters	Accuracy (%)	mAP@0.5 (%)	Time (ms)
Faster-RCNN	110.901 M	53.103 M	71.4	79. 7	33
R-FCN	96.901 M	51.283 M	72.8	79.3	29
YOLOv3	95.827 M	50.853 M	73.5	79.3	27
ResNext-101	235.310 M	112.231 M	73.5	80.5	46
YOLOX	92.437 M	48.569 M	73.6	81.4	21
YOLOv5	89.624 M	46.170 M	73.6	81.8	15
DETR	184.714 M	91.449 M	73.9	82.0	42
Swin-Transformer	131.663 M	69.406 M	74.0	82.2	40

Bold meant the best performance of tampering detection in the same experimental setting

Back to article page