|
|
--- |
|
|
language: |
|
|
- en |
|
|
license: mit |
|
|
size_categories: |
|
|
- 1K<n<10K |
|
|
task_categories: |
|
|
- video-text-to-text |
|
|
--- |
|
|
|
|
|
# $\mathcal{RTV}\text{-}Bench$: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video |
|
|
|
|
|
[](https://arxiv.org/abs/2505.02064) [](https://huggingface.co/datasets/xunsh/RTV-Bench) [](https://www.modelscope.cn/datasets/Jungang/RTV-Bench) |
|
|
[Paper](https://huggingface.co/papers/2505.02064) | [Project Page](https://ljungang.github.io/RTV-Bench) | [Code](https://github.com/ljungang/rtv-bench) |
|
|
|
|
|
## 🔥 News |
|
|
* **`2025-09-20`** 🎉🎉🎉 Our paper has been accepted by NeurIPS 2025, we will update our dataset and code for community as soon as possible~ |
|
|
* **`2025-06-27`** 🎉 We update core code for evaluation. |
|
|
* **`2025-05-17`** 🎉 We have released the label json, which is named `QA.json`. |
|
|
* **`2025-05-04`** 🎉 We released the paper $\mathcal{RTV}\text{-}Bench$: [Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video](https://arxiv.org/abs/2505.02064). |
|
|
* **`2025-05-03`** 🌟 We are happy to release the $\mathcal{RTV}\text{-}Bench$. You can find the $\mathcal{RTV}\text{-}Bench$ from [](https://huggingface.co/datasets/xunsh/RTV-Bench) or [](https://www.modelscope.cn/datasets/Jungang/RTV-Bench). |
|
|
<p align="center"> |
|
|
<img src="https://github.com/ljungang/rtv-bench/blob/main/asset/1_examples.png?raw=true" width="100%" height="100%" > |
|
|
</p> |
|
|
|
|
|
## TODO |
|
|
- [x] Release the final label json. |
|
|
- [x] Release the evaluation code. |
|
|
- [ ] Construct a more comprehensive benchmark for real-time video analysis. |
|
|
- [ ] ··· |
|
|
|
|
|
## 👀 $\mathcal{RTV}\text{-}Bench$ Overview |
|
|
|
|
|
We introduce $\mathcal{RTV}\text{-}Bench$, a fine-grained benchmark for MLLM real-time video analysis, which contains **552** videos (167.2 hours) and **4,631** high-quality QA pairs. We evaluated leading MLLMs, including proprietary (_e.g._ GPT-4o, Gemini 2.0), open-source offline (_e.g._ Qwen2.5-VL, VideoLLaMA3), and open-source real-time (_e.g._ VITA-1.5, InternLM-XComposer2.5-OmniLive) models. Experiment results show open-source real-time models largely outperform offline ones but still trail top proprietary models. Our analysis also reveals that larger model size or higher frame sampling rates do not significantly boost $\mathcal{RTV}\text{-}Bench$ performance, sometimes causing slight decreases. This underscores the need for better model architectures optimized for video stream processing and long sequences to advance real-time video analysis with MLLMs. $\mathcal{RTV}\text{-}Bench$ includes three key principles: |
|
|
* **Multi-Timestamp Question Answering (MTQA)**, where answers evolve with scene changes; |
|
|
* **Hierarchical Question Structure**, combining basic and advanced queries; and |
|
|
* **Multi-dimensional Evaluation**, assessing the ability of continuous perception, understanding, and reasoning. |
|
|
|
|
|
**Video Categories and Distribution of Question Difficulty and Query Characteristics.** |
|
|
<p align="center"> |
|
|
<img src="https://github.com/ljungang/rtv-bench/blob/main/asset/2_dataset_stati.png?raw=true" width="100%" height="100%" > |
|
|
(Left) RTV-Bench overs 3 key domains and 16 sub-class video types. |
|
|
(Center) Distribution of question difficulty levels across eight representative task types, measured by percentage-based performance ranges. |
|
|
(Right) Distribution of question queries by video length, categorized into Shallow, Moderate, and Deep levels. The bar heights indicate counts, while the line chart overlays query proportions for each duration bucket. |
|
|
</p> |
|
|
|
|
|
## 🔖Evaluation Results |
|
|
<p align="center"> |
|
|
<img src="https://github.com/ljungang/rtv-bench/blob/main/asset/3_evaluation.png?raw=true" width="100%" height="100%"> |
|
|
</p> |
|
|
|
|
|
## 🛠️ Sample Usage (Evaluation) |
|
|
To evaluate models on RTV-Bench, you can use the provided script: |
|
|
```shell |
|
|
bash scripts/eval/eval_model.sh |
|
|
``` |
|
|
|
|
|
## 📑 Citation |
|
|
If you find $\mathcal{RTV}\text{-}Bench$ useful for your research and applications, please cite using this BibTeX: |
|
|
```bibtex |
|
|
@article{xun2025rtv, |
|
|
title={RTV-Bench: Benchmarking MLLM Continuous Perception, Understanding and Reasoning through Real-Time Video}, |
|
|
author={Xun, Shuhang and Tao, Sicheng and Li, Jungang and Shi, Yibo and Lin, Zhixin and Zhu, Zhanhui and Yan, Yibo and Li, Hanqian and Zhang, Linghao and Wang, Shikang and others}, |
|
|
journal={arXiv preprint arXiv:2505.02064}, |
|
|
year={2025} |
|
|
} |
|
|
``` |
