Research map / Vision-language and multimodal / Multimodal LLMs
HIVE: Understanding Post Hallucination Reasoning in Vision Language Models
Feng He, Zhenting Wang, Qifan Wang, Qiang Guan, Dongfang Liu, Ruixiang Tang, Qiankun Li
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 |
|---|---|---|---|
| replace faithful caption with hallucinated caption | GPT-4o · PlantVillage | accuracy: 62.73 -> 77.41 | Improved |
| replace faithful caption with hallucinated caption | Qwen-VL-Max · ISIC | accuracy: 58.71 -> 75.62 | Improved |
| replace faithful caption with hallucinated caption | Qwen-VL-Max · GQA | accuracy: 69.88 -> 67.90 | Hurt |
| replace faithful caption with hallucinated caption | GPT-4o, text-only tasks · BBBP | accuracy: 61.67 -> 68.33 | Improved |
| replace faithful caption with hallucinated caption | O3 and DeepSeek R1, text-only tasks · BBBP | accuracy: 73.27 -> 59.47 | Hurt |
| replace faithful caption with hallucinated caption | GPT-4o · CodeXGLUE | accuracy: 55.15 -> 52.75 | No clear effect |
| replace hallucination discriminator with random checker (ablation) | HIVE · ISIC | accuracy: 0.5763 -> 0.5805 | No clear effect |
| mask hallucinated evidence tokens in helpful captions (ablation) | HIVE · AntiCP2 | accuracy: 0.244 | Improved |
| set sampling temperature to 0.6 instead of 0.0 (ablation) | hallucinated caption generation · PlantVillage | accuracy gain (H-F): +2.30 -> +14.68 | Improved |
| raise temperature to 0.9 instead of 0.6 (ablation) | hallucinated caption generation · AntiCP2 | accuracy gain (H-F): +3.76 -> -5.00 | Hurt |
| use 256 token budget instead of 128 (ablation) | hallucinated caption generation · PlantVillage | accuracy gain (H-F): -4.66 -> +14.68 | Improved |
| use 1024 token budget instead of 256 (ablation) | hallucinated caption generation · ISIC | accuracy gain (H-F): +11.76 -> -0.48 | Hurt |
Closest papers by meaning
- More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models (NeurIPS 2025 · similarity 0.94)
- MIRAGE: Assessing Hallucination in Multimodal Reasoning Chains of MLLM (NeurIPS 2025 · similarity 0.94)
- Auditing Meta-Cognitive Hallucinations in Reasoning Large Language Models (NeurIPS 2025 · similarity 0.94)
- Combating Multimodal LLM Hallucination via Bottom-Up Holistic Reasoning (AAAI 2025 · similarity 0.93)
- Multi-Object Hallucination in Vision Language Models (NeurIPS 2024 · similarity 0.92)
- HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Models (CVPR 2024 · 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.
