Symmetric Entropy-Constrained Video Coding for Machines

Based on the DCVC-HEM baseline architecture (ACM MM 2022), extended with our proposed modules (bi-directional entropy-constraint and semantic-pixel dual-path fusion).

Cases Visualziation

Model Weights

The pre-trained model weights are not publicly available at this moment, as they are tied to ongoing projects, patents, and commercial usage.

Requirements

• Python 3.8+
• PyTorch >= 2.0.0
• See requirements.txt for full list

Acknowledgement

This project builds upon:

• DCVC-HEM — Base video codec architecture
• CompressAI — Learned compression primitives
• Mask2Former — Semantic teacher model
• DINOv2 — Visual foundation model
• Detectron2 — Detection framework

📝 Citation

If you find this work useful for your research, please cite:

@article{sun2026secvcm,
  title   = {Symmetric Entropy-Constrained Video Coding for Machines},
  author  = {Sun, Yuxiao and Liu, Meiqin and Yao, Chao and Tang, Qi and Jin, Jian and Lin, Weisi and Dufaux, Frederic and Zhao, Yao},
  journal = {IEEE Transactions on Image Processing},
  year    = {2026},
  doi     = {10.1109/TIP.2026.3705185}
}

📄 License

This project is released under the MIT License. Third-party components retain their original licenses.