AI Paper Atlas

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

CVPR 2026 · Video › Video-language understanding · approach: Long video understanding

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 changedWhereMeasuredEffect
DPO with in-video hallucination pairs (4k) (ablation)Qwen2.5-VL-7B · ELV-HallucSAH Ratio: 8.3 -> 6.0Improved
DPO with out-of-video hallucination pairs (4k) (ablation)Qwen2.5-VL-7B · ELV-HallucSAH Ratio: 8.3 -> 8.6Hurt
DPO with combined in-video and out-of-video pairs (8k) (ablation)Qwen2.5-VL-7B · ELV-HallucSAH Ratio: 8.3 -> 8.4No clear effect
DPO with in-video pairs (4k) (ablation)Qwen2.5-VL-7B · VideoMMEaverage accuracy: 61.9 -> 62.3Improved
DPO with out-of-video pairs (4k) (ablation)Qwen2.5-VL-7B · VideoMMEaverage accuracy: 61.9 -> 62.8Improved
use VideoRoPE instead of vanilla RoPE (ablation)Qwen2-VL checkpoints · ELV-HallucSAH Ratio: 1.82 -> 0.88Improved
use TAD-RoPE instead of vanilla RoPE (ablation)Qwen2-VL checkpoints · ELV-HallucSAH Ratio: 1.82 -> 2.18Hurt
use stronger RoPE variants (m-RoPE, VideoRoPE) for overall hallucination (ablation)Qwen2-VL checkpoints · ELV-HallucOut-video accuracy: 2.75 -> 2.06No clear effect
increase number of sampled frames (ablation)Qwen2.5-VL and InternVL3 · ELV-HallucSAH RatioHurt
increase number of sampled frames (ablation)Qwen2.5-VL and InternVL3 · ELV-Hallucoverall hallucination accuracyImproved
scale up LLM size (ablation)Video-MLLMs · ELV-HallucSAH RatioNo clear effect
scale up LLM size (ablation)Video-MLLMs · ELV-Hallucoverall hallucination accuracyImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy