SV3.3B - Sports Video Description Model

⚠️ RESEARCH USE ONLY - NON-COMMERCIAL LICENSE

SV3.3B is a unified video-to-text model specifically designed for sports video understanding.

License

This model is licensed under CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0 International). This model is intended for research purposes only and cannot be used for commercial applications.

Model Architecture

• Vision Encoder: Custom DWT-VJEPA2-based video encoder
• Text Model: 3.3B parameter language model
• Vision-Text Bridge: Learned projection layer
• Specialization: Fine-tuned on sports video data

Model Details

• Parameters: 3.3B
• Input: Video files (16 frames, 256x256 resolution)
• Output: Natural language descriptions
• Domain: Sports video analysis
• Training: SPORTSVISION/NSVA_SUBSET

Model Performance

Limitations

• Research use only
• No commercial applications
• Optimized for sports content
• May not generalize to other video domains

Citation

If you use this model in your research, please cite:

@misc{sv3-3b-2024,
  title={SV3.3B: Sports Video Description Model},
  author={Varun Kodathala},
  year={2025},
  url={https://huggingface.co/sportsvision/SV3.3B}
}