SynShot - Synthetic Prior for Few-Shot Drivable Head Avatar Inversion

Wojciech Zielonka^1,2,3*, Stephan J. Garbin³, Alexandros Lattas³,
George Kopanas³, Paulo Gotardo³, Thabo Beeler³, Justus Thies^1,2, Timo Bolkart³

¹Max Planck Institute for Intelligent Systems, Tübingen, Germany
²Technical University of Darmstadt ³Google

📺 Video  📄 Paper  🌐 Website 

Official Repository for CVPR 2025 paper SynShot - Synthetic Prior for Few-Shot Drivable Head Avatar Inversion

Ablation and comparison to Synshot

In order to compare to our method we release videos for GAN baselines and personalized monocular methods.

The videos were created using the INSTA dataset, with the last 600 frames designated as test sequences.

To evaluate your method:

1. Download the dataset.
2. Train your model while excluding the last 600 frames.
3. Use these 600 frames for testing and comparison with SynShot.

For cross-reenactment videos, in addition to transferring expressions, we also transfer the camera parameters from the source sequence.

Personalized baselines

Video Order (from left to right):

| Source/Target | Ours | INSTA | FlashAvatar | Splatting Avatar |

🟢 Explore Results: (all results were generated using 3 input frames for inversion for our method)
📥 Monocular Test Sequences
📥 Monocular Transfer Results

🎭 Expression Transfer & Novel View Results (we release all the results for different numer of input frames ranging from 3 to 13)
📥 Inversion Output per Actor

📷 Download Input Frames:
📥 Input Frames

GAN-based baselines

Video Order (from left to right):

| Source/Target | Ours | InvertAvatar | Portrait4D | Next3D |

📥 Explore GAN-BASELINES

Citation

If you use this project in your research please cite SynShot:

@inproceedings{zielonka2025synshot,
    title={Synthetic Prior for Few-Shot Drivable Head Avatar Inversion},
    author={Wojciech Zielonka and Stephan J. Garbin and Alexandros Lattas and George Kopanas and Paulo Gotardo and Thabo Beeler and Justus Thies and Timo Bolkart},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month = {June},
    year={2025},
}

*Work done while Wojciech Zielonka was interning at Google Zurich