notesum.ai

Published at December 6

Beyond Boxes: Mask-Guided Spatio-Temporal Feature Aggregation for Video Object Detection

cs.CV

Released Date: December 6, 2024

Authors: Khurram Azeem Hashmi¹, Talha Uddin Sheikh¹, Didier Stricker¹, Muhammad Zeshan Afzal¹

Aff.: ¹DFKI - German Research Center for Artificial Intelligence

Arxiv: http://arxiv.org/pdf/2412.04915v1

Refer to caption

Method	Source	Backbone	mAP(%) $\uparrow$	Time (ms) $\downarrow$
SELSA [65]	ICCV2019	X101	83.1	153.8
RDN [16]	ICCV2019	R101	81.8	162.6
MEGA [5]	CVPR2020	R101	82.9	230.4
TROIA [22]	AAAI2021	X101	84.3	285.7
MAMBA [55]	AAAI2021	R101	84.6	110.3(T)
QueryProp [28]	AAAI2022	R101	82.3	30.8(T)
SparseVOD [27]	BMVC2022	R101	81.9	142.4
FAQ [9]	CVPR2023	R50	81.7	163.2
Liu et al. [43]	ICCV2023	R101	87.2	39.6(T)
STPN [56]	ICCV2023	SwinT	85.0	45.7
TransVODLite [71]	TPAMI2022	SwinT	83.7	42.1
YOLOV-S [54]	AAAI2023	MCSP	77.3	11.3
YOLOV-L [54]		MCSP	83.6	16.3
YOLOV-X [54]		MCSP	85.0	22.7
FAIM-S	Ours	MCSP	78.2_+0.9	11.6
FAIM-L		MCSP	84.3_+0.7	16.5
FAIM-X		MCSP	85.6_+0.6	22.7
With Post-processing
YOLOV-S [54]	AAAI2023	MCSP	80.1	11.3 $+$ 6.9
YOLOV-L [54]		MCSP	86.2	16.3 $+$ 6.9
YOLOV-X [54]		MCSP	87.2	22.7 $+$ 6.1
FAIM-S	Ours	MCSP	80.6_+0.5	11.6 $+$ 6.9
FAIM-L		MCSP	87.0_+0.8	16.5 $+$ 6.9
FAIM-X		MCSP	87.9_+0.7	22.7 $+$ 6.9