Victoria Fernández Abrevaya*, Adnane Boukhayma*, Philip H. S. Torr, Edmond Boyer (*Equal contrib.).
CVPR 2020 (Oral)

• Python 2.7
• PyTorch 0.3

Input images are assumed to be crops of fixed size around the face. Using dlib, this command finds the tightest rectangular box of edge size l containing the face. Images are then cropped with a square patch of size 1.2xl. Input images are located in data/original and cropped images are saved in data/cropped.

Download the dlib trained facial shape predictor. Put file shape_predictor_68_face_landmarks.dat in directory data.

python crop.py

Download the model weights. Put file model.pth in directory data.

Run the following command to generate an image of normals from a cropped RGB image example in data/cropped. Results are saved in data/output.

python tester.py

@InProceedings{Abrevaya_2020_CVPR,
author = {Abrevaya, Victoria Fernandez and Boukhayma, Adnane and Torr, Philip H.S. and Boyer, Edmond},
title = {Cross-Modal Deep Face Normals With Deactivable Skip Connections},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2020}
} 

This work was partly supported by the ERC grant ERC-2012-AdG 321162-HELIOS, the EPSRC grant Seebibyte EP/M013774/1 and the EPSRC/MURI grant EP/N019474/1.

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.