Released on 04/25/2022

This is the implementation of Shizhan Zhu et al.'s ICLR-22 work Differentiable Gradient Sampling for Learning Implicit 3D Scene Reconstruction from a Single Image. It is open source under the MIT license (see the LICENSE file). Codes can be used freely for any ethical purpose.

DGS is among the first attempts to tackle the generalizable 3D implicit surface reconstruction from a single image for indoor scenes, via directly learning from raw scan data.

1. Installations, training and testing for ShapeNet.
2. Installations, training and testing for Scannet, and demo for arbitrary inputs.

If you use the codes as part of your research project, please cite our work as follows:

@inproceedings{zhu2021differentiable,
  title={Differentiable Gradient Sampling for Learning Implicit 3D Scene Reconstructions from a Single Image},
  author={Zhu, Shizhan and Ebrahimi, Sayna and Kanazawa, Angjoo and Darrell, Trevor},
  booktitle={International Conference on Learning Representations},
  year={2021}
}

Suggestions and opinions of this work (both positive and negative) are greatly welcome. Please contact the author by sending email to zhshzhutah2@gmail.com.

MIT, see LICENSE file for details.