search

unibas-gravis / Occlusion-Robust-MoFA

Other
208 stars 30 forks source link

train/val/test_landmarks_3D.csv of CelebA-HQ #1

Closed kennyvoo closed 3 years ago

kennyvoo commented 3 years ago

Can you release the train/val/test_landmarks_3D.csv of CelebA-HQ ?

kennyvoo commented 3 years ago

I've tried to prepare the landmarks_3d.csv for CelebA-HQ using the 2D-and-3D-face-alignment repo.

  1. So do I produce a 3d landmark or 2d landmark? In the load_dataset.py line 83, the landmark is reshaped into tensors of shape So, it's a 2d landmarks instead of 3d?
    • landmark = torch.Tensor(landmark_cpu).reshape(-1,2).transpose(0,1)
jo-dean commented 3 years ago

So did you konw it's 3d landmark or 2d landmark? thank you?

ChunLLee commented 3 years ago

2D landmarks are used in our implementation. Please use FAN to predict 68 2D landmarks.