PointTransformer: Encoding Human Local Features for Small Target Detection

Table 4

Comparison with the SOTA model.

Method	AP	AP50	AP75	APs	APm	APl
FCOS	28.1	39.5	31.3	16.5	35.7	48.1
EfficentNet-B0-based EfficientDet	27.9	38.6	30.8	16.8	34.5	46.6
EfficentNet-B3-based EfficientDet	30.8	40.8	33.6	17.0	36.2	49.1
Transformer-based Deformable DETR	33.5	42.7	36.1	19.4	41.5	51.7
YOLOV5-X	31.4	40.1	33.2	17.3	38.7	53.5
Our proposed model	37.2	45.5	40.7	29.3	42.1	43.2