Skip to yearly menu bar Skip to main content


HalluciDoctor: Mitigating Hallucinatory Toxicity in Visual Instruction Data

Qifan Yu · Juncheng Li · Longhui Wei · Liang Pang · Wentao Ye · Bosheng Qin · Siliang Tang · Qi Tian · Yueting Zhuang

Arch 4A-E Poster #320
[ ]
Thu 20 Jun 10:30 a.m. PDT — noon PDT


Multi-modal Large Language Models (MLLMs) tuned on machine-generated instruction-following data have demonstrated remarkable performance in various multi-modal understanding and generation tasks. However, the hallucinations inherent in machine-generated data, which could lead to hallucinatory outputs in MLLMs, remain under-explored. This work aims to investigate various hallucinations (i.e., object, relation, attribute hallucinations) and mitigate those hallucinatory toxicities in large-scale machine-generated visual instruction datasets. Drawing on the human ability to identify factual errors, we present a novel hallucination detection and elimination framework, HalluciDoctor, based on the cross-checking paradigm. We use our framework to identify and eliminate hallucinations in the training data automatically. Interestingly, HalluciDoctor also indicates that spurious correlations arising from long-tail object co-occurrences contribute to hallucinations. Based on that, we execute counterfactual visual instruction expansion to balance data distribution, thereby enhancing MLLMs' resistance to hallucinations. Comprehensive experiments on hallucination evaluation benchmarks show that our method successfully mitigates 44.6% hallucinations relatively and maintains competitive performance compared to LLaVA. The data and code for this paper are publicly available.

Live content is unavailable. Log in and register to view live content