tencent-ailab / hifi3dface

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

How to use augmented 3dmm bases for single image input #21

Open Barry-Liang opened 3 years ago

Barry-Liang commented 3 years ago

I noticed that the run_RGB_opt with a single input image uses the no-augmented bases. The output geometry is a full head, which doesn't require another step to add the full head. And there is no difference between males and females about the shoulder and other geometries. Therefore, I try to modify the bash file to use augmented 3dmm bases and use the add_head tool in RGBD utils to make a full head for single image input. The final result for the face region is better than the previous but the backhead is too long and strange. Do you have any idea about how to solve it or where should I change?

olgaiv39 commented 3 years ago

Hi @Barry-Liang, Can you please share how did you solve an issue with compiling tf_mesh_renderer here? Especially, I am interested in your working versions of Cuda, Cudnn, Tensorflow and Bazel. Thank you in advance!