Research map / Vision-language and multimodal / Multimodal LLMs
AVTrustBench: Assessing and Enhancing Reliability and Robustness in Audio-Visual LLMs
Sanjoy Chowdhury, Sayan Nag, Subhrajyoti Dasgupta, Yaoting Wang, Mohamed Elhoseiny, Ruohan Gao, Dinesh Manocha
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 DPO with CAVPref (audio-visual conditioned, robust preference optimization) | VideoLLaMA2 · AVTrustBench MVIT | accuracy: 36.53 -> 54.83 | Improved |
| replace DPO with CAVPref | video-SALMONN · AVTrustBench MAIT | accuracy: 35.18 -> 54.77 | Improved |
| add the distributionally robust module to CAVPref (ablation) | CAVPref on VideoLLaMA2 · AVTrustBench MCIT | accuracy: 36.11 -> 41.45 | Improved |
| add the distributionally robust module to CAVPref (ablation) | CAVPref on VideoLLaMA2 · AVTrustBench MAT | accuracy: 64.77 -> 69.12 | Improved |
| add the distributionally robust module to CAVPref (ablation) | CAVPref on video-SALMONN · AVTrustBench MAIT | accuracy: 49.12 -> 54.77 | Improved |
| add audio-visual conditioning terms (L^V, L^A) beyond DPO (ablation) | VideoLLaMA2 · AVTrustBench MAIT | accuracy: 32.16 -> 46.51 | Improved |
| use DPO instead of SFT | VideoLLaMA2 · AVTrustBench MAIT | accuracy: 31.18 -> 32.16 | No clear effect |
| use DPO instead of SFT | VideoLLaMA2 · AVTrustBench MCIT | accuracy: 25.68 -> 35.82 | Improved |
| set lambda_y and lambda_V to 1.0 (ablation) | CAVPref · AVTrustBench MCIT, MVIT | accuracy | Improved |
| set lambda_A to 0.8 (ablation) | CAVPref · AVTrustBench MAIT | accuracy | Improved |
| scale up the LLM backbone | AVLLMs · AVTrustBench compositional reasoning | accuracy | No clear effect |
| use ImageBind pre-aligned encoders | AVLLMs · AVTrustBench zero-shot | accuracy | Improved |
Closest papers by meaning
- HAVE-Bench: Hierarchical Audio-Visual Evaluation from Perception to Interaction (CVPR 2026 · similarity 0.91)
- AVI-Bench: Toward Human-like Audio-Visual Intelligence of Omni-MLLMs (ICML 2026 · similarity 0.91)
- When Eyes and Ears Disagree: Can MLLMs Discern Audio-Visual Confusion? (AAAI 2026 · similarity 0.90)
- AVHBench: A Cross-Modal Hallucination Benchmark for Audio-Visual Large Language Models (ICLR 2025 · similarity 0.90)
- JointAVBench: A Benchmark for Joint Audio-Visual Reasoning Evaluation (ICLR 2026 · similarity 0.89)
- $\texttt{AVROBUSTBENCH}$: Benchmarking the Robustness of Audio-Visual Recognition Models at Test-Time (NeurIPS 2025 · similarity 0.89)
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.
