Visually Dehallucinative Instruction Generation: Know What You Don't Know

Sungguk Cha, Jusung Lee, Younghyun Lee, Cheoljong Yang
2/15/2024
cs.CV

No official code found. Best implementation: ncsoft/idk (6 stars, BSD-3-Clause, last commit 2024-02-16).

Abstract

"When did the emperor Napoleon invented iPhone?" Such hallucination-inducing question is well known challenge in generative language modeling. In this study, we present an innovative concept of visual hallucination, referred to as "I Know (IK)" hallucination, to address scenarios where "I Don't Know" is the desired response. To effectively tackle this issue, we propose the VQAv2-IDK benchmark, the subset of VQAv2 comprising unanswerable image-question pairs as determined by human annotators. Stepping further, we present the visually dehallucinative instruction generation method for IK hallucination and introduce the IDK-Instructions visual instruction database. Our experiments show that current methods struggle with IK hallucination. Yet, our approach effectively reduces these hallucinations, proving its versatility across different frameworks and datasets.

View on arXivView PDF

Code Implementations(1)

ncsoft/idkCommunity100%
BSD-3-Clause

Official implementation of "Visually Dehallucinative Instruction Generation: Know What You Don't Know"

62Feb 14, 20242 years agoBSD-3-Clause

Similar Papers

CIEM: Contrastive Instruction Evaluation Method for Better Instruction Tuning

90% match2023

Mitigating Low-Level Visual Hallucinations Requires Self-Awareness: Database, Model and Training Strategy

88% match2025

LongHalQA: Long-Context Hallucination Evaluation for MultiModal Large Language Models

88% match1 repo52024

Investigating and Mitigating the Multimodal Hallucination Snowballing in Large Vision-Language Models

87% match1 repo182024

Mitigating Object Hallucination in MLLMs via Data-augmented Phrase-level Alignment

87% match2024

Negative Object Presence Evaluation (NOPE) to Measure Object Hallucination in Vision-Language Models

87% match2023

Cite this paper

@article{cha2024visually,
  title  = {Visually Dehallucinative Instruction Generation: Know What You Don't Know},
  author = {Sungguk Cha and Jusung Lee and Younghyun Lee and Cheoljong Yang},
  year   = {2024},
  eprint = {2402.09717},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2402.09717v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2402.09717.svg)](https://codeofpaper.com/paper/2402.09717)

Discussion