tencent-ailab / hifi3dface

Code and data for our paper "High-Fidelity 3D Digital Human Creation from RGB-D Selfies".
Other
774 stars 153 forks source link

Why 3d landmark detect is still needed thus we already have depth data? #26

Open LiuLinyun opened 3 years ago

LiuLinyun commented 3 years ago

Why 3d landmark detect is still needed thus we already have depth data?

In the paper only mentioned "2D landmarks for each frame are lifted from 2D to 3D using depth data" but not mentioned using 3D landmark detector model to detect 86 3D-landmarks, can only use depth data to map 2D landmarks to 3D by bfs depth search in file "optimization/rgbd/RGBD_utils/SparseFusion.py" function "bfs_find_corrospondence"?