Object Detection Benchmark | The MM-AU Benchmark Suite

Object Detection Evaluation

This is the 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.

Leaderboard

Detectors	Years	V1-Train [V_o, V_r, V_a]						V2-Train [V_o, V_r]				Anchor	GFlops	#Params.
		val. [V_o, V_r, V_a]		test. [V_o, V_r, V_a]		test. [V_a]		test. [V_o, V_r]		test. [V_a]
		mAP50	AR	mAP50	AR	mAP50	AR	mAP50	AR	mAP50	AR
FasterRCNN	2015	0.674	0.634	0.666	0.623	0.664	0.620	0.544	0.524	0.497	0.509	✓	0.19T	41.38M
CornerNet	2018	0.495	0.625	0.485	0.619	0.483	0.624	0.436	0.563	0.456	0.598		0.71T	201M
CascadeRPN	2019	0.662	0.699	0.664	0.696	0.649	0.689	0.579	0.663	0.532	0.624	✓	0.18T	41.97M
CenterNet	2019	0.054	0.238	0.051	0.233	0.047	0.224	0.161	0.260	0.155	0.257		20.38G	14.21M
DeTR	2020	0.367	0.407	0.377	0.403	0.363	0.403	0.275	0.329	0.254	0.318		44.55G	28.83M
EfficientNet	2020	0.310	0.417	0.310	0.412	0.293	0.404	0.075	0.128	0.073	0.133		57.28G	18.46M
Deformable-DETR	2021	0.660	0.671	0.661	0.668	0.652	0.663	0.626	0.631	0.587	0.626		0.18T	40.1M
YOLOx	2021	0.673	0.709	0.672	0.698	0.670	0.698	0.563	0.627	0.540	0.626		13.33G	8.94M
YOLOv5s	2021	0.757	0.766	0.748	0.764	0.743	0.761	0.660	0.716	0.636	0.712	✓	8.13G	12.35M
DiffusionDet	2023	0.731	0.749	0.733	0.745	0.718	0.738	0.701	0.729	0.660	0.716		-	26.82M
YOLOv8	2023	0.716	0.754	0.715	0.753	0.717	0.755	0.606	0.702	0.597	0.703		14.28G	11.14M
YOLOv11	2025	0.698	0.684	0.696	0.690	0.693	0.692	0.673	0.662	0.639	0.649		21.6G	9.43M
YOLO26	2025	0.716	0.752	0.713	0.758	0.710	0.760	0.682	0.708	0.646	0.695		22.5G	9.95M

Submit Your Results

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

Click here to submit your results:

Submit