Research map / Vision-language and multimodal / Multimodal LLMs
ORBIT: Overcoming Hallucination Risks via Bi-manifold Interaction and Traction
Zhehan Kan, Yanlin Liu, Xiaochen Yang, Hongyang Yu, Qingmin Liao, Wenming Yang
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 ORBIT intervention | LLaVA-v1.5-7B · POPE / CHAIR-S | CHAIR-S: 51.2 -> 38.1; POPE F1 85.3 -> 87.5 | Improved |
| add ORBIT intervention | Qwen2.5-VL-7B · CHAIR-S | CHAIR-S: 26.5 -> 18.4 | Improved |
| add AGLA baseline | Qwen2.5-VL-7B · POPE | F1: 88.4 -> 87.7 | Hurt |
| add GSTI-triggered adaptive intervention versus always-on correction (ablation) | ORBIT · LLaVA-v1.5-7B POPE/CHAIR | CHAIR-S: 41.3 -> 38.1 | Improved |
| include residual extraction step (ablation) | ORBIT · LLaVA-v1.5-7B CHAIR-S | CHAIR-S: 45.8 -> 38.1 | Improved |
| include visual re-anchoring (ablation) | ORBIT · LLaVA-v1.5-7B CHAIR-S | CHAIR-S: 44.2 -> 38.1 | Improved |
| include iso-norm injection (ablation) | ORBIT · LLaVA-v1.5-7B CHAIR-S | CHAIR-S: 42.7 -> 38.1 | Improved |
| include visual deviation signal in GSTI (ablation) | ORBIT · LLaVA-v1.5-7B CHAIR-S | CHAIR-S: 40.7 -> 38.1 | Improved |
| include semantic volatility signal in GSTI (ablation) | ORBIT · LLaVA-v1.5-7B CHAIR-S | CHAIR-S: 39.8 -> 38.1 | Improved |
| increase context window K from 3 to 20 (ablation) | ORBIT · LLaVA-v1.5-7B POPE | POPE: 87.2 -> 87.8; latency 5.60 -> 13.22 | No clear effect |
| add ORBIT on general benchmarks | LLaVA-v1.5-7B · MME / MMBench / CV-Bench / CLEVR | accuracy: MME 1495.5 -> 1525.8; CLEVR 43.6 -> 46.4 | Improved |
| add ORBIT to LLaVA-v1.5-7B on GEOBench-VLM | LLaVA-v1.5-7B · GEOBench-VLM | accuracy: 25.1 -> 26.1 | Improved |
Closest papers by meaning
- Reducing Hallucinations in Large Vision-Language Models via Latent Space Steering (ICLR 2025 · similarity 0.91)
- HallusionBench: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Models (CVPR 2024 · similarity 0.91)
- Causally-Grounded Dual-Path Attention Intervention for Object Hallucination Mitigation in LVLMs (AAAI 2026 · similarity 0.91)
- Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models (CVPR 2026 · similarity 0.90)
- The Hidden Life of Tokens: Reducing Hallucination of Large Vision-Language Models Via Visual Information Steering (ICML 2025 · similarity 0.90)
- Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lens (CVPR 2025 · similarity 0.90)
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.
