Closed risingsunho closed 6 years ago
You are referring to VRN - Multitask, which produces 2D landmarks for the image, while also regressing the 3D reconstruction. This model is not publicly available. Besides, because the method is volumetric, the landmarks would not be in correspondence with the volume.
I read about your wonderful paper. And I runned your code and it works perfect. In figure 4-c, the proposed VRN regresses both the 3D facial volume and a set of sparse facial landmarks. Can I get 3D model's facial landmark from this code?