notesum.ai

Published at December 6

DEYOLO: Dual-Feature-Enhancement YOLO for Cross-Modality Object Detection

cs.CV

Released Date: December 6, 2024

Authors: Yishuo Chen¹, Boran Wang, Xinyu Guo, Wenbin Zhu, Jiasheng He, Xiaobin Liu, Jing Yuan

Aff.: ¹Nankai University

Arxiv: http://arxiv.org/pdf/2412.04931v1

Refer to caption

Method	Modality	mAP₅₀	mAP_50-95
Swin Transformer [18]	visible	76.4	44.9
	infrared	72.6	41.9
	cross-modality	73.8	42.6
CenterNet2 [36]	visible	78.5	52.4
	infrared	65.3	42.4
	cross-modality	70.2	46.5
Sparse RCNN [23]	visible	82.4	49.6
	infrared	76.4	44.8
	cross-modality	78.2	47.3
YOLOv7-tiny [29]	visible	82.1	51.6
	infrared	78.1	48.4
	cross-modality	80.1	49.8
YOLOv7 [29]	visible	90.4	61.3
	infrared	87.9	58.3
	cross-modality	88.3	59.6
YOLOv8n [12]	visible	80.8	54.3
	infrared	78.3	52.3
	cross-modality	79.2	52.8
YOLOv8l [12]	visible	88.3	61.8
YOLOv8l [12]	infrared	86.5	59.6
DEYOLO-n(ours)	Cross-modality	86.6	58.9
DEYOLO-l(ours)	Cross-modality	$\mathbf{91.2}$	$\mathbf{66.3}$