Research map / Vision-language and multimodal / Multimodal LLMs
Learning to Mask: Cross-Modal Noise Modulation for Hallucination Mitigation in Multi-modal Large Language Models
Zijian Song, Chunlei Wang, Kun He
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Image Noise Modulation Module (In-Mod) (ablation) | LLaVA-1.5-7B · CHAIR (MSCOCO, 64 tokens) | CHAIRs: 21.20 -> 12.80 | Improved |
| add Text Noise Modulation Module (Tn-Mod) (ablation) | LLaVA-1.5-7B · CHAIR (MSCOCO, 64 tokens) | CHAIRs: 21.20 -> 20.40 | Improved |
| add Tn-Mod on top of In-Mod (ablation) | LLaVA-1.5-7B · CHAIR (MSCOCO, 64 tokens) | CHAIRs: 12.80 -> 7.60 | Improved |
| add In-Mod to MiniGPT-4 (ablation) | MiniGPT-4-7B · CHAIR (MSCOCO, 64 tokens) | CHAIRs: 32.40 -> 20.80 | Improved |
| add Tn-Mod to MiniGPT-4 (ablation) | MiniGPT-4-7B · CHAIR (MSCOCO, 64 tokens) | CHAIRs: 32.40 -> 23.00 | Improved |
| add both modulation modules to MiniGPT-4 (ablation) | MiniGPT-4-7B · CHAIR (MSCOCO, 64 tokens) | CHAIRs: 32.40 -> 16.20 | Improved |
| add dynamical regularization term (ablation) | HalMask LLaVA-1.5-7B · POPE (average) | F1: 87.11 -> 87.52 | Improved |
| add dynamical regularization term, recall (ablation) | HalMask LLaVA-1.5-7B · POPE (average) | recall: 82.76 -> 84.98 | Improved |
| add dynamical regularization term, RTD (ablation) | HalMask LLaVA-1.5-7B · POPE (average) | RTD (lower better): 56.13 -> 50.15 | Improved |
| apply HalMask to LLaVA-1.5 on CHAIR | LLaVA-1.5-7B · CHAIR (MSCOCO, 64 tokens) | BLEU: 16.25 -> 19.48 | Improved |
| apply HalMask to Qwen2.5-VL | Qwen2.5-VL-7B · CHAIR (MSCOCO, 512 tokens) | CHAIRs: 36.1 -> 23.7 | Improved |
| apply HalMask to LLaVA-1.5 on AMBER Cover | LLaVA-1.5-7B · AMBER | Cover: 50.9 -> 50.1 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Nullu — Nullu: Mitigating Object Hallucinations in Large Vision-Language Models via HalluSpace Projection (CVPR 2025) · better on 4 of 4 shared results
- HALC — HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding (ICML 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Reducing Hallucinations in Large Vision-Language Models via Latent Space Steering (ICLR 2025 · similarity 0.94)
- Grounding Language with Vision: A Conditional Mutual Information Calibrated Decoding Strategy for Reducing Hallucinations in LVLMs (NeurIPS 2025 · similarity 0.94)
- Self-Introspective Decoding: Alleviating Hallucinations for Large Vision-Language Models (ICLR 2025 · similarity 0.94)
- Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs (AAAI 2026 · similarity 0.94)
- Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation (ICLR 2026 · similarity 0.93)
- Residual Decoding: Mitigating Hallucinations in Large Vision-Language Models via History-Aware Residual Guidance (CVPR 2026 · similarity 0.93)
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.
