notesum.ai

Published at December 10

Exploring What Why and How: A Multifaceted Benchmark for Causation Understanding of Video Anomaly

cs.CV

cs.AI

Released Date: December 10, 2024

Authors: Hang Du¹, Guoshun Nan¹, Jiawen Qian¹, Wangchenhui Wu¹, Wendi Deng¹, Hanqing Mu¹, Zhenyan Chen¹, Pengxuan Mao², Xiaofeng Tao¹, Jun Liu³

Aff.: ¹Beijing University of Posts and Telecommunications; ²Terminus Technologies Co., Ltd.; ³Lancaster University

Arxiv: http://arxiv.org/pdf/2412.07183v1

Refer to caption

Metrics	Answer Pool Ranking
Metrics	Description	Cause	Effect
ROUGE [37]	11.1%	5.7%	10.2%
BLEU [36]	9.2%	3.4%	13.1%
BLEURT [59]	15.3%	15.6%	22.4%
MoverScore [64]	40.1%	45.9%	38.7%
UniEval [63]	42.3%	42.9%	50.4%
GPT-Base [63]	71.3%	73.2%	69.7%
AnomEval (Ours)	82.3%	80.2%	89.1%