VNBench

We utilize [VideoNIAH](https://github.com/joez17/VideoNIAH) to compile a video benchmark **VNBench**, including tasks such as **retrieval**, **ordering**, and **counting**. **VNBench** contains 1350 samples in total.
**VNBench** can efficiently evaluate the fine-grained understanding ability and spatio-temporal modeling ability of a video model, while also supporting the long-context evaluation.

We use 4-try circular evaluation for VNBench. Thus the whole annotation file contains 5400 samples.

我们依托[VideoNIAH](https://github.com/joez17/VideoNIAH)构建了视频基准测试集**VNBench**，该测试集涵盖**检索**、**排序**与**计数**三类任务。**VNBench**总计包含1350个样本。

**VNBench**可高效评估视频模型的细粒度理解能力及时空建模能力，同时支持长上下文评估任务。

我们针对VNBench采用4轮循环评测机制，因此完整标注文件共计包含5400个样本。