Research map / Vision-language and multimodal / Multimodal LLMs
Mitigating Object Hallucinations via Sentence-Level Early Intervention
Shangpin Peng, Senqiao Yang, Li Jiang, Zhuotao Tian
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 |
|---|---|---|---|
| use context-aware DPO instead of standard DPO (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 10.1 -> 4.3 | Improved |
| use non-hallucinated context instead of hallucinated context (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 14.3 -> 4.3 | Improved |
| use non-hallucinated context instead of natural context (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 8.6 -> 4.3 | Improved |
| add iterative contextual bootstrapping (ICB) (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 5.3 -> 4.3 | Improved |
| add ICB (ablation) | SENTINEL · MM-Vet | overall: 31.8 -> 32.6 | Improved |
| use original in-domain pairs instead of rewritten pairs (ablation) | SENTINEL · AMBER | accuracy: 75.0 -> 76.1 | Improved |
| ignore uncertain objects instead of treating as factual (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 10.3 -> 4.3 | Improved |
| ignore uncertain objects instead of treating as hallucinated (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 8.3 -> 4.3 | Improved |
| use two detectors (Grounding DINO + YOLO World) instead of one (ablation) | SENTINEL detector · Object HalBench | CHAIR response-level: 14.3 -> 6.6 | Improved |
| use only positive-free mixed data (50% negative-positive) (ablation) | SENTINEL · Object HalBench | CHAIR response-level: 4.3 -> 4.8 | Hurt |
| add SENTINEL data to HA-DPO | HA-DPO · Object HalBench | CHAIR response-level: 37.0 -> 8.0 | Improved |
| SENTINEL versus base model on general VQA | LLaVA-v1.5-7B · TextVQA | accuracy: 58.2 -> 58.2 | No clear effect |
Closest papers by meaning
- Detecting and Mitigating Hallucination in Large Vision Language Models via Fine-Grained AI Feedback (AAAI 2025 · similarity 0.93)
- Mitigating Multimodal Hallucinations via Gradient-based Self-Reflection (CVPR 2026 · similarity 0.92)
- Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models (CVPR 2026 · similarity 0.92)
- Stop Learning it all to Mitigate Visual Hallucination, Focus on the Hallucination Target. (CVPR 2025 · similarity 0.91)
- Mitigating Object Hallucinations in Large Vision-Language Models with Assembly of Global and Local Attention (CVPR 2025 · similarity 0.91)
- Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations (ICLR 2025 · similarity 0.91)
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.
