VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912669412360192 |
|---|---|
| author | Li, Zongxia Wu, Xiyang Shi, Guangyao Qin, Yubin Du, Hongyang Liu, Fuxiao Zhou, Tianyi Manocha, Dinesh Boyd-Graber, Jordan Lee |
| author_facet | Li, Zongxia Wu, Xiyang Shi, Guangyao Qin, Yubin Du, Hongyang Liu, Fuxiao Zhou, Tianyi Manocha, Dinesh Boyd-Graber, Jordan Lee |
| contents | Vision-Language Models (VLMs) have achieved strong results in video understanding, yet a key question remains: do they truly comprehend visual content or only learn shallow correlations between vision and language? Real visual understanding, especially of physics and common sense, is essential for AI systems that interact with the physical world. Current evaluations mostly use real-world videos similar to training data, so high benchmark scores may not reflect real reasoning ability. To address this, we propose negative-control tests using videos that depict physically impossible or logically inconsistent events. We introduce VideoHallu, a synthetic dataset of physics- and commonsense-violating scenes generated with Veo2, Sora, and Kling. It includes expert-annotated question-answer pairs across four categories of violations. Tests of leading VLMs (Qwen-2.5-VL, Video-R1, VideoChat-R1) show that, despite strong results on benchmarks such as MVBench and MMVU, they often miss these violations, exposing gaps in visual reasoning. Reinforcement learning fine-tuning on VideoHallu improves recognition of such violations without reducing standard benchmark performance. Our data is available at https://github.com/zli12321/VideoHallu.git. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_01481 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding Li, Zongxia Wu, Xiyang Shi, Guangyao Qin, Yubin Du, Hongyang Liu, Fuxiao Zhou, Tianyi Manocha, Dinesh Boyd-Graber, Jordan Lee Computer Vision and Pattern Recognition Machine Learning Vision-Language Models (VLMs) have achieved strong results in video understanding, yet a key question remains: do they truly comprehend visual content or only learn shallow correlations between vision and language? Real visual understanding, especially of physics and common sense, is essential for AI systems that interact with the physical world. Current evaluations mostly use real-world videos similar to training data, so high benchmark scores may not reflect real reasoning ability. To address this, we propose negative-control tests using videos that depict physically impossible or logically inconsistent events. We introduce VideoHallu, a synthetic dataset of physics- and commonsense-violating scenes generated with Veo2, Sora, and Kling. It includes expert-annotated question-answer pairs across four categories of violations. Tests of leading VLMs (Qwen-2.5-VL, Video-R1, VideoChat-R1) show that, despite strong results on benchmarks such as MVBench and MMVU, they often miss these violations, exposing gaps in visual reasoning. Reinforcement learning fine-tuning on VideoHallu improves recognition of such violations without reducing standard benchmark performance. Our data is available at https://github.com/zli12321/VideoHallu.git. |
| title | VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2505.01481 |