notesum.ai

Published at November 27

Dual-Level Boost Network for Long-Tail Prohibited Items Detection in X-ray Security Inspection

cs.CV

Released Date: November 27, 2024

Authors: Renshuai Tao¹, Haoyu Wang¹, Wei Wang¹, Yunchao Wei¹, Yao Zhao¹

Aff.: ¹Beijing Jiaotong University

Arxiv: http://arxiv.org/abs/2411.18078v1

Refer to caption

Method	HiXray Dataset									PIDXray Dataset
Method	AP₅₀	PO1	PO2	WA	LA	MP	TA	CO	NL (Tail)	AP₅₀	BA	PL	HA	PO	SC	WR	GU	BU	SP	HA	KN	LI
F-RCNN [26]	84.2	95.0	93.4	91.9	98.0	97.0	95.7	71.0	32.0	76.0	83.7	96.0	88.5	60.7	91.2	96.7	47.6	71.9	67.3	98.7	42.1	67.8
+ ours	85.2	96.1	94.4	92.4	98.0	97.0	94.1	68.0	41.7_↑9.7%	77.6	87.7	96.4	90.0	63.2	92.5	97.0	48.2	72.6	72.8	98.7	42.8	69.6
S-RCNN[31]	80.8	94.2	92.4	90.3	98.2	97.0	95.1	63.6	15.1	71.8	84.4	95.0	86.7	62.3	86.4	96.8	37.1	66.0	48.6	98.5	33.9	65.7
+ ours	83.2	94.8	93.6	91.2	97.9	97.3	94.5	66.5	29.7_↑14.6%	72.4	85.0	94.5	83.3	63.5	86.3	96.3	38.5	66.8	56.0	98.6	34.3	66.5
RetinaNet[27]	79.0	94.0	93.2	90.8	98.1	97.3	95.4	61.8	1.7	69.0	77.6	93.1	82.8	48.9	83.7	91.6	38.4	64.2	58.2	98.0	28.0	63.5
+ ours	81.8	94.8	93.8	91.0	97.9	97.5	95.1	65.6	18.9 ${}_{\color[rgb]{0.8,0,0}{\uparrow\textbf{17.2}\%}}$	70.5	77.8	93.8	84.9	48.8	84.8	92.6	47.3	63.6	58.1	98.0	31.9	64.4
CenterNet[9]	82.4	95.6	94.3	90.7	98.1	97.8	95.3	73.1	14.6	74.0	86.5	96.8	90.3	56.9	90.6	95.6	35.3	68.7	67.9	98.5	35.3	65.7
+ ours	83.7	95.5	94.5	90.9	98.0	97.9	95.4	73.2	24.3_↑9.7%	74.3	87.0	96.9	91.4	54.7	90.3	97.0	36.9	72.8	66.7	98.4	33.3	65.8
ATSS[45]	83.5	95.2	93.6	91.8	98.2	97.4	95.4	68.0	28.3	74.6	84.1	96.3	91.2	56.5	90.9	96.4	39.8	70.6	71.4	98.5	33.0	66.7
+ ours	85.6	95.7	94.2	93.1	98.1	97.4	95.4	70.6	40.7_↑12.4%	76.2	87.6	96.9	90.0	58.2	90.0	96.6	49.3	72.1	74.7	98.0	32.8	67.8