This page contains end-to-end demo code to recover accurate and detailed reconstructions of the facial shape from an unconstrained 2D face image. For a given input image, it fits a 3D Morphable Face Model [2] to the face image and then uses a CGAN to refine the reconstruction by adding local fine-grained details. The code outputs a standard obj file (without texture). The full approach is described in our paper [1].

• cmake
• dlib
• face-alignment
• opencv-python
• scipy==1.1.0
• scikit-image
• tensorflow

Dependencies can be installed by running pip install -r requirements.txt

Use the following link to download the 3DMM data and the CGAN weights.

The data3dmm folder and the .pb file must be placed in the root folder.

The script can be run with the following: python main.py < --im_path imgname > < --use_camera >

If --im_path is not specified, the default image is used. Test images must be placed into testdata folder.

If a webcam is available, --use_camera can be used to capture a live frame.

If you find this work useful, please cite

@article{galteri2019deep,
  title={Deep 3D morphable model refinement via progressive growing of conditional Generative Adversarial Networks},
  author={Galteri, Leonardo and Ferrari, Claudio and Lisanti, Giuseppe and Berretti, Stefano and Del Bimbo, Alberto},
  journal={Computer Vision and Image Understanding},
  volume={185},
  pages={31--42},
  year={2019},
  publisher={Elsevier}
}

[1] Galteri, Leonardo; Ferrari, Claudio, et al. "Deep 3D morphable model refinement via progressive growing of conditional Generative Adversarial Networks." Computer Vision and Image Understanding 185 (2019): 31-42.

[2] Ferrari, Claudio, et al. "A dictionary learning-based 3D morphable shape model." IEEE Transactions on Multimedia 19.12 (2017): 2666-2679.

For any further question, please write to [email protected] or [email protected]