Research map / Video / Video-language understanding
ELV-Halluc: Benchmarking Semantic Aggregation Hallucinations in Video Understanding
Hao Lu, Jiahao Wang, Yaolun Zhang, Ruohui Wang, Xuanyu Zheng, Yepeng Tang, Dahua Lin, Lewei Lu
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 |
|---|---|---|---|
| DPO with in-video hallucination pairs (4k) (ablation) | Qwen2.5-VL-7B · ELV-Halluc | SAH Ratio: 8.3 -> 6.0 | Improved |
| DPO with out-of-video hallucination pairs (4k) (ablation) | Qwen2.5-VL-7B · ELV-Halluc | SAH Ratio: 8.3 -> 8.6 | Hurt |
| DPO with combined in-video and out-of-video pairs (8k) (ablation) | Qwen2.5-VL-7B · ELV-Halluc | SAH Ratio: 8.3 -> 8.4 | No clear effect |
| DPO with in-video pairs (4k) (ablation) | Qwen2.5-VL-7B · VideoMME | average accuracy: 61.9 -> 62.3 | Improved |
| DPO with out-of-video pairs (4k) (ablation) | Qwen2.5-VL-7B · VideoMME | average accuracy: 61.9 -> 62.8 | Improved |
| use VideoRoPE instead of vanilla RoPE (ablation) | Qwen2-VL checkpoints · ELV-Halluc | SAH Ratio: 1.82 -> 0.88 | Improved |
| use TAD-RoPE instead of vanilla RoPE (ablation) | Qwen2-VL checkpoints · ELV-Halluc | SAH Ratio: 1.82 -> 2.18 | Hurt |
| use stronger RoPE variants (m-RoPE, VideoRoPE) for overall hallucination (ablation) | Qwen2-VL checkpoints · ELV-Halluc | Out-video accuracy: 2.75 -> 2.06 | No clear effect |
| increase number of sampled frames (ablation) | Qwen2.5-VL and InternVL3 · ELV-Halluc | SAH Ratio | Hurt |
| increase number of sampled frames (ablation) | Qwen2.5-VL and InternVL3 · ELV-Halluc | overall hallucination accuracy | Improved |
| scale up LLM size (ablation) | Video-MLLMs · ELV-Halluc | SAH Ratio | No clear effect |
| scale up LLM size (ablation) | Video-MLLMs · ELV-Halluc | overall hallucination accuracy | Improved |
Closest papers by meaning
- VidHalluc: Evaluating Temporal Hallucinations in Multimodal Large Language Models for Video Understanding (CVPR 2025 · similarity 0.92)
- Learning to Decode Against Compositional Hallucination in Video Multimodal Large Language Models (ICML 2026 · similarity 0.91)
- Mitigating Hallucination in VideoLLMs via Temporal-Aware Activation Engineering (NeurIPS 2025 · similarity 0.91)
- MASH-VLM: Mitigating Action-Scene Hallucination in Video-LLMs through Disentangled Spatial-Temporal Representations (CVPR 2025 · similarity 0.91)
- Analyzing and Mitigating Object Hallucination: A Training Bias Perspective (AAAI 2026 · similarity 0.90)
- SVHalluc: Benchmarking Speech-Vision Hallucination in Audio-Visual Large Language Models (CVPR 2026 · 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.
