notesum.ai

Published at December 5

DEIM: DETR with Improved Matching for Fast Convergence

cs.CV

cs.AI

Released Date: December 5, 2024

Authors: Shihua Huang¹, Zhichao Lu, Xiaodong Cun, Yongjun Yu, Xiao Zhou, Xi Shen

Aff.: ¹Intellindust AI Lab

Arxiv: http://arxiv.org/pdf/2412.04234v1

Refer to caption

Model	#Epochs	#Params	GFLOPs	Latency (ms)	AP^val	AP ${}^{val}_{50}$	AP ${}^{val}_{75}$	AP ${}^{val}_{S}$	AP ${}^{val}_{M}$	AP ${}^{val}_{L}$
YOLO-based Real-time Object Detectors
YOLOv8-L [12]	500	43	165	12.31	52.9	69.8	57.5	35.3	58.3	69.8
YOLOv8-X [12]	500	68	257	16.59	53.9	71.0	58.7	35.7	59.3	70.7
YOLOv9-C [34]	500	25	102	10.66	53.0	70.2	57.8	36.2	58.5	69.3
YOLOv9-E [34]	500	57	189	20.53	55.6	72.8	60.6	40.2	61.0	71.4
Gold-YOLO-L [33]	300	75	152	9.21	53.3	70.9	-	33.8	58.9	69.9
YOLOv10-L^⋆ [32]	500	24	120	7.66	53.2	70.1	58.1	35.8	58.5	69.4
YOLOv10-X^⋆ [32]	500	30	160	10.74	54.4	71.3	59.3	37.0	59.8	70.9
YOLO11-L^⋆ [13]	500	25	87	6.31	52.9	69.4	57.7	35.2	58.7	68.8
YOLO11-X^⋆ [13]	500	57	195	10.52	54.1	70.8	58.9	37.0	59.2	69.7
DETR-based Real-time Object Detectors
RT-DETR-HG-L [43]	72	32	107	8.77	53.0	71.7	57.3	34.6	57.4	71.2
RT-DETR-HG-X [43]	72	67	234	13.51	54.8	73.1	59.4	35.7	59.6	72.9
D-FINE-L [27]	72	31	91	8.07	54.0	71.6	58.4	36.5	58.0	71.9
DEIM-D-FINE-L	50	31	91	8.07	54.7	72.4	59.4	36.9	59.6	71.8
D-FINE-X [27]	72	62	202	12.89	55.8	73.7	60.2	37.3	60.5	73.4
DEIM-D-FINE-X	50	62	202	12.89	56.5	74.0	61.5	38.8	61.4	74.2