AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Beyond Blind Noising: Disentangled Visual Rectification for Hallucination Mitigation in MLLMs

Yujia Chen, Rui Sun, Bingzhou Wang, Huayu Mai, Wangkai Li, Zhaoyang Li, Aibing Li, Wenzhang SUN

ICML 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Decoding-time hallucination mitigation

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Shows that noise-induced drift used by Visual Contrastive Decoding entangles structural degradation with hallucination-relevant signal, and proposes a disentangled visual rectification method to mitigate hallucinations in MLLMs.

Problem

Hallucinations in multimodal large language models; existing Visual Contrastive Decoding assumes the output shift from noise-perturbed images isolates the hallucination direction, which the paper argues is flawed for Language-Image Pretrained encoders.

Approach

Argues that noise-induced drift in LIP encoders is a coupled vector mixing structural degradation and hallucination signal, and disentangles these components to rectify visual representations rather than penalizing blind noise-based contrast.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add DVR decodingLLaVA-1.5-7B · CHAIRCHAIRs: 54.2 -> 47.4Improved
add DVR decodingLLaVA-1.5-7B · CHAIRRecall: 72.5 -> 78.6Improved
use DVR instead of VCDLLaVA-1.5-7B · POPE Adversarialaccuracy: 78.2 -> 80.2Improved
add VCD to Qwen3-VL-8BQwen3-VL-8B · CHAIRCHAIRs: 21.0 -> 23.7Hurt
add DVR to Qwen3-VL-8BQwen3-VL-8B · CHAIRCHAIRs: 21.0 -> 19.0Improved
add Stream I with DINOv3 auxiliary encoder (without Stream II) (ablation)DVR · CHAIRCHAIRs: 54.2 -> 49.2Improved
use DINOv3 instead of DINOv2 as auxiliary encoder (ablation)DVR Stream I · CHAIRCHAIRs: 50.4 -> 49.2Improved
use SigLIP (LIP paradigm) as auxiliary encoder instead of SSL (ablation)DVR Stream I · CHAIRCHAIRs: 54.2 -> 53.6No clear effect
add Stream II decoding contrast to Stream I (ablation)DVR · CHAIRCHAIRs: 49.2 -> 47.4Improved
add Stream I rectification to VCD-style contrast (ablation)DVR · CHAIRCHAIRs: 51.8 -> 47.4Improved
scale DVR from 7B to 13B (ablation)DVR on LLaVA-1.5 · CHAIRCHAIRs: 47.4 -> 45.4Improved
use DVR (7B) versus cross-scale VCD (13B-7B) (ablation)LLaVA-1.5-7B · CHAIRCHAIRs: 47.0 vs 47.4No clear effect

Surprising findings

Experimental setup

Baselines: Visual Contrastive Decoding

Builds on: Visual Contrastive Decoding

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy