notesum.ai

Published at December 6

MANTA: A Large-Scale Multi-View and Visual-Text Anomaly Detection Dataset for Tiny Objects

cs.CV

Released Date: December 6, 2024

Authors: Lei Fan¹, Dongdong Fan², Zhiguang Hu³, Yiwen Ding², Donglin Di⁴, Kai Yi⁵, Maurice Pagnucco¹, Yang Song¹

Aff.: ¹UNSW Sydney; ²Gaozhe; ³SCAU; ⁴Li Auto; ⁵University of Cambridge

Arxiv: http://arxiv.org/pdf/2412.04867v1

Refer to caption

Model(’year)	Agriculture		Medicine		Electronics		Mechanics		Groceries		Average
	view-eval	object-eval	view-eval	object-eval	view-eval	object-eval	view-eval	object-eval	view-eval	object-eval	view-eval	object-eval
RD’22 [11]	84.7/85.2	87.2/85.3	86.4/94.0	92.3/94.1	86.6/93.5	89.2/93.4	81.0/91.9	85.7/91.8	71.2/85.3	73.5/85.3	82.0/90.0	85.6/90.0
PatchCore’22 [42]	93.0/93.0	94.2/93.0	96.5/96.7	97.4/97.0	97.1/98.8	97.0/98.8	95.8/98.7	96.4/98.7	86.2/91.5	90.8/91.5	93.7/95.7	95.2/95.8
CDO’23 [8]	90.0/90.6	91.3/90.4	95.2/95.4	95.2/95.5	94.1/98.6	91.9/98.6	91.1/98.4	90.3/98.4	85.4/90.4	87.7/89.3	91.2/94.7	91.3/94.4
DMAD’23 [31]	84.5/84.9	86.9/85.2	84.6/93.3	91.2/91.2	87.4/92.1	90.1/92.2	80.2/90.3	85.3/90.4	75.1/87.5	77.0/87.4	82.6/89.6	86.1/89.7
SimpleNet’23 [33]	86.3/83.3	91.2/84.3	91.1/87.2	94.7/88.3	89.6/90.3	92.5/91.2	86.4/88.7	89.6/88.5	82.2/82.5	86.5/84.9	87.1/86.4	90.9/87.5