Datasets:

---

Viglong

/

DSI-Bench

like 1

This PR significantly enhances the dataset card for DSI-Bench by adding crucial information for users. Specifically, it includes:

• The task_categories: ['video-text-to-text'] to the metadata, improving discoverability.
• Relevant tags such as video-question-answering, video-understanding, spatial-reasoning, benchmark, and 3d.
• A link to the official Hugging Face paper: https://huggingface.co/papers/2510.18873
• Direct links to the project page (https://dsibench.github.io/) and the GitHub repository (https://github.com/SpatialVision/dsibench).
• The paper abstract for a concise overview of the dataset.
• A comprehensive "Sample Usage" section, featuring code snippets for installation, data download, inference with the Qwen API, and evaluation of model performance, all directly from the GitHub README.
• The BibTeX citation for referencing the work.

These additions provide a much more comprehensive and user-friendly resource for researchers interested in Dynamic Spatial Intelligence.