Research map / Vision-language and multimodal / Cross-modal learning
OmniDPO: A Preference Optimization Framework to Address Omni-Modal Hallucination
Junzhe Chen, Tianshu Zhang, Shiyu Huang, Yuwei Niu, Chao Sun, Rongzhou Zhang, Guanyu Zhou, Lijie Wen
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 modality-aware visual and audio preference losses on top of text DPO | Qwen2.5-Omni-7B · AVHBench Audio-driven Video Hallucination | accuracy: 71.74 -> 84.42 | Improved |
| add text-only DPO | Qwen2.5-Omni-7B · AVHBench Audio-driven Video Hallucination | accuracy: 74.12 -> 71.74 | Hurt |
| add VCD contrastive decoding | Qwen2.5-Omni-7B · AVHBench Audio-driven Video Hallucination | accuracy: 74.12 -> 77.40 | Improved |
| add text-only DPO | MiniCPM-o-2.6 · CMM overall | hallucination resistance (hr): 63.9 -> 66.2 | Improved |
| add OmniDPO | MiniCPM-o-2.6 · CMM overall | hallucination resistance (hr): 63.9 -> 68.8 | Improved |
| add audio preference loss only (no visual loss) (ablation) | OmniDPO on Qwen2.5-Omni-7B · AVHBench Video-driven Audio Hallucination | accuracy: 67.6 -> 76.7 | Improved |
| add video preference loss only (no audio loss) (ablation) | OmniDPO on Qwen2.5-Omni-7B · CMM Audio-Language (AL) | hallucination resistance (hr): 78.0 -> 63.0 | Hurt |
| add video preference loss only (no audio loss) (ablation) | OmniDPO on Qwen2.5-Omni-7B · CMM Visual-Audio-Language (VAL) | hallucination resistance (hr): 90.5 -> 66.0 | Hurt |
| combine audio and video preference losses (ablation) | OmniDPO on Qwen2.5-Omni-7B · CMM Audio-Language (AL) | hallucination resistance (hr): 82.5 (audio only) -> 83.5 | Improved |
| add video preference loss only (ablation) | OmniDPO on Qwen2.5-Omni-7B · AVHBench Audio-driven Video Hallucination | accuracy: 74.1 -> 82.4 | Improved |
| add OmniDPO training | Qwen2.5-Omni-7B · AVHBench Audio-driven Video Hallucination | Yes rate (%): 64.44 -> 44.28 | Improved |
| add OmniDPO training | Qwen2.5-Omni-7B and MiniCPM-o-2.6 · MMAU and MMMU (zero-shot) | accuracy: average +2.4% and +3.6% | Improved |
Later papers that compared with this method
2 later accepted papers report OmniDPO as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Don't Let the Video Speak: Audio-Contrastive Preference Optimization for Audio-visual Language Models (ECCV 2026 · better on 6 of 6 shared results)
- MoD-DPO: Towards Mitigating Cross-modal Hallucinations in Omni LLMs using Modality Decoupled Preference Optimization (CVPR 2026 · better on 5 of 5 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
- MoD-DPO: Towards Mitigating Cross-modal Hallucinations in Omni LLMs using Modality Decoupled Preference Optimization (CVPR 2026 · similarity 0.94)
- AVHBench: A Cross-Modal Hallucination Benchmark for Audio-Visual Large Language Models (ICLR 2025 · similarity 0.92)
- The Curse of Multi-Modalities: Evaluating Hallucinations of Large Multimodal Models across Language, Visual, and Audio (NeurIPS 2025 · similarity 0.92)
- Learning to Decode Against Compositional Hallucination in Video Multimodal Large Language Models (ICML 2026 · similarity 0.92)
- Combating Multimodal LLM Hallucination via Bottom-Up Holistic Reasoning (AAAI 2025 · similarity 0.92)
- Robust Multimodal Large Language Models Against Modality Conflict (ICML 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.
