Research map / Video / Video-language understanding
ARGUS: Hallucination and Omission Evaluation in Video-LLMs
Ruchit Rawal, Reza Shirkavand, Heng Huang, Gowthami Somepalli, Tom Goldstein
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 |
|---|---|---|---|
| apply DPO post-training to LLaVA-NeXT-Video (ablation) | LLaVA-NeXT-Video 7B · ARGUS | ArgusCost-H (%): 59 -> 45 | Improved |
| increase model scale within a family (SmolVLM2 256M to 2B) (ablation) | SmolVLM2 · ARGUS | ArgusCost-H: over 15% hallucination improvement | Improved |
| increase model size within InternVL2 family (ablation) | InternVL2 · ARGUS | ArgusCost-H and ArgusCost-O | No clear effect |
| use larger Gemini-2.0-Pro instead of Gemini-2.0-Flash (ablation) | Gemini · ARGUS | ArgusCost-H (%): 41 -> 48 | Hurt |
| increase frames from 2 to 64 for Gemini-2.0-Flash (ablation) | Gemini-2.0-Flash · ARGUS 50-video subset | ArgusCost-H: improves by 10.02% | Improved |
| increase frames from 2 to 64 for MiniCPM-V-2.6 and Qwen2.5-VL (ablation) | MiniCPM-V-2.6, Qwen2.5-VL · ARGUS 50-video subset | ArgusCost-H | Mixed |
| increase frames from 2 to 64 (ablation) | all tested models · ARGUS 50-video subset | ArgusCost-O: Gemini-2.0-Flash improves by 9.69% | Improved |
| use longer generated captions | Video-LLMs · ARGUS | ArgusCost-O correlation with caption length: 0.09 (omission), 0.32 (hallucination) | No clear effect |
| swap GPT-4o judge for DeepSeek-R1, DeepSeek-V3, LLaMa-3.3 or Qwen-2.5 (ablation) | ARGUS evaluation · model ranking | Pearson ranking correlation: 0.96, 0.97, 0.93, 0.92 | No clear effect |
Closest papers by meaning
- VISTA-LLAMA: Reducing Hallucination in Video Language Models via Equal Distance to Visual Tokens (CVPR 2024 · similarity 0.89)
- Toward Robust Hyper-Detailed Image Captioning: A Multiagent Approach and Dual Evaluation Metrics for Factuality and Coverage (ICML 2025 · similarity 0.89)
- SVHalluc: Benchmarking Speech-Vision Hallucination in Audio-Visual Large Language Models (CVPR 2026 · similarity 0.88)
- VidHalluc: Evaluating Temporal Hallucinations in Multimodal Large Language Models for Video Understanding (CVPR 2025 · similarity 0.88)
- When Semantics Mislead Vision: Mitigating Large Multimodal Models Hallucinations in Scene Text Spotting and Understanding (NeurIPS 2025 · similarity 0.88)
- Mitigating Hallucination in VideoLLMs via Temporal-Aware Activation Engineering (NeurIPS 2025 · similarity 0.88)
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.
