Document

Object Detection Evaluation

This is 2D object detection benchmark. We annotate over 2.23M object boxes for about 463K video frames. The benchmark uses 2D bounding box overlapping rate to compute precision-recall curves for calculating the AP, mAP75, AP50, mAP75, AP_S, AP_M, and AP_L metrics in various pedestrian categories.

NOTE: You can submit your metric values via the provided form. Furthermore, we would highly appreciate for your contribution with clear links to relevant articles and code for more in-depth analysis.

the result please submit here:

ID	Method	AP	AP50	mAP75	AP_S	AP_M	AP_L	Runtime	Environment	Code	Paper

ID	Method	Year	Code	AP	AP50	mAP75	AP_S	AP_M	AP_L	Runtime	Environment
1	FasterRCNN	2015	Code	45.4 %	71.5 %	55.2 %	49.1 %	47.3 %	37.6 %	...	GeForce RTX 3090
2	CornerNet	2018	Code	37.8 %	54.9 %	43.9 %	31.7 %	43.6 %	25.2 %	...	GeForce RTX 3090
3	CascadeRPN	2019	Code	44.8 %	69.9 %	51.3 %	44.3 %	46.0 %	42.4 %	...	GeForce RTX 3090
4	CenterNet	2019	Code	1.1 %	4.0 %	0.2 %	3.7 %	1.4 %	1.1 %	...	GeForce RTX 3090
5	DETR	2020	Code	9.9 %	29.4 %	3.8 %	9.6 %	8.8 %	1.5 %	...	GeForce RTX 3090
6	EfficientNet	2021	Code	11.4 %	29.9 %	5.5 %	9.6 %	12.7 %	10.6 %	...	GeForce RTX 3090
7	Deformable-DeTR	2020	Code	40.4 %	68.6 %	44.1 %	39.6 %	42.1 %	36.1 %	...	GeForce RTX 3090
8	YOLOx	2021	Code	42.4 %	69.5 %	47.1 %	38.7 %	44.0 %	40.6 %	...	GeForce RTX 3090
9	YOLOv5s	2021	Code	52.9 %	78.4 %	63.2 %	45.9 %	54.4 %	42.1 %	...	GeForce RTX 3090
10	DiffusionDet	2023	Code	52.7 %	76.7 %	60.7 %	46.3 %	54.4 %	51.6 %	...	GeForce RTX 3090
11	YOLOv8	2023	Code	50.6 %	74.8 %	59.0 %	45.5 %	51.6 %	51.2 %	...	GeForce RTX 3090