VideoVista-2

# VideoVista This repository contains VideoVista2, a versatile benchmark for multicultural, multilingual, and multidomain video comprehension. We develop a scalable, quality-controlled workflow that first uses an automatic QA generation framework to produce a large volume of candidate QA pairs from curated, domain-diverse videos. ## 🔥 News [2026/03/01] 🔥 We have open-sourced the answers of VideoVista2 benchmark! [2026/01/18] 🔥 VideoVista2 is currently released as the Benchmark B for the [VideoVista Competition](https://github.com/HITsz-TMG/VideoVista/tree/main/VideoVista-Competition), so only the original video files and questions are provided. The answers will be released on 2026/02/08. ## 🌟 Files We provide the questions in both 'test-00000-of-00001.parquet' and 'VideoVista2_no_answer.json' files. To unzip the videos and images, use the following code. ```shell zip -s 0 videos.zip --out videos_full.zip unzip videos_full.zip unzip images.zip ``` ## Evaluation We use `Accuracy` to evaluate the performance of VideoVista. ## Experimental Results  ## Citation If you find VideoVista useful for your research and applications, please cite using this BibTeX: ```bibtex @inproceedings{chen2025videovista, title={VideoVista-CulturalLingo: 360^{\circ} Horizons-Bridging Cultures, Languages, and Domains in Video Comprehension}, author={Chen, Xinyu and Li, Yunxin and Shi, Haoyuan and Hu, Baotian and Luo, Wenhan and Wang, Yaowei and Zhang, Min}, booktitle={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)}, year={2025}, publisher={Association for Computational Linguistics}, address={Vienna, Austria}, pages={27102--27128} } @article{li2024videovista, title={Videovista: A versatile benchmark for video understanding and reasoning}, author={Li, Yunxin and Chen, Xinyu and Hu, Baotian and Wang, Longyue and Shi, Haoyuan and Zhang, Min}, journal={arXiv preprint arXiv:2406.11303}, year={2024} } ```