A vanilla 3D face modeling on pose-invariant and multi-lightning image data

• Background
• Install
• Usage
• Contributing

The project tries to restore surface normal and albedo information based on pose-invariant and multi-lightning image data, then use them to rebuild the 3d face model.

Test Data

Recovered Surface

This project is implemented with Google Colab. Please download all the files and upload them into your own Google Drive project folder and then open it.

First, load images

# load images as intensity matrix at each pixel
imstack = []
for i in range(1, 8):
  imstack.append((cv2.imread(Images path, cv2.IMREAD_GRAYSCALE)).flatten())
imstack = np.vstack(imstack)

Second, define data light

# load light source vector S
S = pickle.load(open('./data/sources.pickle', 'rb'))['S']

The project's recover surface function is credited to Todd Zickler, CS283, Harvard University.