kumuji

/

Sa2VA-i-1B

like 0

This PR significantly enhances the model card for this Sa2VA-i model by:

• Updating the pipeline_tag to image-segmentation. This change more accurately reflects the model's core functionality of language-guided dense grounding and referring video object segmentation, ensuring better discoverability on the Hugging Face Hub.
• Adding explicit links to the Hugging Face paper page (Sa2VA-i: Improving Sa2VA Results with Consistent Training and Inference) and the official GitHub repository (https://github.com/kumuji/sa2va-i).
• Expanding the model description with a detailed overview, performance highlights, competition results, model zoo, quick start guidance, and key improvements, all sourced directly from the project's GitHub README. This provides comprehensive information to users.
• Including authors and a teaser image for better presentation.
• Consolidating citations to include both the Sa2VA-i paper and the original Sa2VA paper.
• Removing irrelevant "File information" from the public model card.

These updates aim to provide a more complete, accurate, and user-friendly model card.