Research map / Vision-language and multimodal / Multimodal LLMs
Paying More Attention to Images: A Training-Free Method for Alleviating Hallucination in LVLMs
Shi Liu, Kecheng Zheng, Wei Chen
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 |
|---|---|---|---|
| amplify image-token attention and subtract text-only logits at inference (PAI) | LLaVA greedy decoding · CHAIR | CHAIR_S: 46.6 -> 24.8 | Improved |
| apply PAI with nucleus sampling | LLaVA nucleus sampling · CHAIR | CHAIR_S: 58.2 -> 43.4 | Improved |
| PAI on Minigpt4 nucleus sampling POPE accuracy vs VCD | Minigpt4 nucleus sampling · POPE single-turn | accuracy: VCD 59.97 vs PAI 59.19 | No clear effect |
| add layer prior restricting intervention layers (ablation) | PAI on Shikra · CHAIR | CHAIR_S: 56.4 -> 37.2 | Improved |
| add layer prior restricting intervention layers (ablation) | PAI on LLaVA · CHAIR | F1: 71.7 -> 75.7 | Improved |
| increase alpha too far (ablation) | PAI on LLaVA · CHAIR | F1: 75.9 -> 62.6 at alpha 0.6 | Hurt |
| increase gamma beyond 1.2 (ablation) | PAI on Minigpt4 · CHAIR | F1: 70.3 -> 64.3 at gamma 2.0 | Hurt |
| uniform attention bias on all image tokens (Prompt Highlighter style) instead of PAI proportional amplification (ablation) | LLaVA-1.5-7B · CHAIR | CHAIR_S: 23.4 -> 52.8 | Hurt |
| apply PAI to larger LLaVA-1.5 13B model | LLaVA-1.5-13B · CHAIR | CHAIR_S: 44.0 -> 33.0 | Improved |
| apply PAI on MMHal-Bench logical questions (comparison, relations) | LVLMs · MMHal-Bench | score | No clear effect |
Closest papers by meaning
- Reducing Hallucinations in Large Vision-Language Models via Latent Space Steering (ICLR 2025 · similarity 0.93)
- AdaIAT: Adaptively Increasing Attention to Generated Text to Alleviate Hallucinations in LVLM (CVPR 2026 · similarity 0.93)
- Analyzing and Mitigating Object Hallucination: A Training Bias Perspective (AAAI 2026 · similarity 0.93)
- Grounding Language with Vision: A Conditional Mutual Information Calibrated Decoding Strategy for Reducing Hallucinations in LVLMs (NeurIPS 2025 · similarity 0.92)
- Self-Correcting Decoding with Generative Feedback for Mitigating Hallucinations in Large Vision-Language Models (ICLR 2025 · similarity 0.92)
- Self-Introspective Decoding: Alleviating Hallucinations for Large Vision-Language Models (ICLR 2025 · similarity 0.92)
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.
