This PR addresses the following points to improve the dataset card:
- Adds `task_categories: ['image-segmentation', 'object-detection', 'video-text-to-text']` to the metadata, making the dataset discoverable through these tasks on the Hub.
- Adds `license: cc-by-nc-sa-4.0` to the metadata for clear licensing information.
- Integrates the Hugging Face paper link (`https://huggingface.co/papers/2512.10945`) into the main navigation links and updates the inline paper reference for the TPAMI 2025 paper.
- Removes the duplicated and verbose paper abstract from the content, adhering to the guidelines for concise dataset cards.
- Removes the redundant license statement from the content, as it's now properly defined in the metadata.
