notesum.ai

Published at November 19

Visual-Oriented Fine-Grained Knowledge Editing for MultiModal Large Language Models

cs.CV

cs.AI

Released Date: November 19, 2024

Authors: Zhen Zeng¹, Leijiang Gu, Xun Yang², Zhangling Duan, Zenglin Shi¹, Meng Wang

Aff.: ¹Hefei University of Technology, Hefei, Anhui, China; ²University of Science and Technology of China, Hefei, Anhui, China

Arxiv: http://arxiv.org/abs/2411.12790v1

	BLIP-2 OPT				MiniGPT-4
	Reliability	Locality	Generality	Specificity	Reliability	Locality	Generality	Specificity
FT-LLM	100.0	76.93	99.96	24.21	93.39	86.26	93.35	35.02
FT-Visual	99.68	100.0	\ul99.21	16.56	93.39	100.0	91.87	30.53
IKE	\ul99.89	48.49	98.02	20.07	100.0	52.45	98.88	25.26
SERAC	93.08	\ul99.90	96.83	31.92	\ul99.50	100.0	92.90	37.85
MSCKE	99.13	100.0	98.56	61.60	\ul99.50	100.0	93.00	57.20
MEND	97.00	98.60	96.40	\ul65.85	94.85	\ul98.58	94.82	\ul67.39
MSCKE-MEND	97.40	100.0	96.50	68.38	97.05	100.0	\ul96.70	71.98