notesum.ai

Published at December 5

HyperDefect-YOLO: Enhance YOLO with HyperGraph Computation for Industrial Defect Detection

cs.CV

Released Date: December 5, 2024

Authors: Zuo Zuo¹, Jiahao Dong², Yue Gao³, Zongze Wu⁴

Aff.: ¹Xi'an Jiaotong University; ²Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ); ³Tsinghua University; ⁴Shenzhen University

Arxiv: http://arxiv.org/pdf/2412.03969v1

Method	Pre	mAP50	Missing_hole		Mouse_bite		Open_circuit		Short		Spur		Spurious_copper
Method	Pre	mAP50	mAP50	Pre	mAP50	Pre	mAP50	Pre	mAP50	Pre	mAP50	Pre	mAP50	Pre
YOLO-v7 [39]	86.5	88.4	89.6	84.8	87.9	86.5	87.4	81.1	86.2	88.7	90.4	88.2	89.1	89.8
YOLO-v8 [8]	87.3	90.1	90.1	85.5	88.1	87.3	88.9	83.2	90.9	89.7	91.2	87.6	91.6	90.3
YOLO-v10 [36]	89.7	91.5	98.9	94.8	88.7	95.0	96.9	89.6	90.1	82.3	96.0	93.3	78.4	83.1
YOLO-v11 [37]	96.7	96.0	99.5	98.4	95.7	98.0	99.3	97.8	96.4	90.3	93.6	100	91.4	95.7
RT-DETR [10]	87.3	88.6	89.1	85.1	89.3	89.7	86.0	85.9	88.4	89.0	90.1	86.5	88.7	87.4
DetectoRS [40]	90.4	93.9	98.4	97.3	97.2	96.8	87.9	86.5	93.8	81.0	91.9	90.4	94.1	90.6
PCGAN [41]	87.6	88.7	88.1	90.6	87.7	86.9	88.5	84.6	88.3	87.8	89.2	85.9	90.1	89.9
ES-Net [42]	94.1	96.3	97.5	96.1	96.3	96.9	98.9	97.3	95.4	92.1	94.7	88.6	94.7	93.6
DeYOLOX [43]	89.5	89.3	88.5	88.1	88.3	89.8	89.9	91.3	88.9	89.3	90.3	90.8	89.6	87.9
CA-FPN [44]	93.4	94.7	96.5	94.3	95.6	96.5	97.4	96.3	95.2	92.8	93.7	89.1	89.5	91.4
CCEANN [45]	93.5	94.0	97.8	95.5	97.3	98.0	96.5	92.7	93.6	92.4	94.6	89.6	84.1	92.5
MSFTI-FDet [46]	94.7	96.3	98.6	97.1	96.3	98.3	97.0	96.5	95.8	93.0	95.5	90.1	94.8	93.4
DRFA [15]	96.0	97.9	99.6	98.2	98.7	98.6	98.5	98.3	97.6	94.5	96.3	91.6	96.4	95.0
HyperYOLO [28]	95.5	96.1	99.5	99.2	98.7	98.3	97.9	97.6	91.1	90.7	97.6	92.7	91.6	94.2
HD-YOLO (ours)	98.1	98.2	99.5	100	98.0	100	99.3	96.4	99.2	97.3	94.1	98.0	98.9	97.0