vrn

Why we included this project

vrn is the reference implementation of the Volumetric Regression Network, which turns a single 2D photo of a face into a full 3D mesh. Rather than predicting landmarks or a sparse point set, the network frames reconstruction as a segmentation problem: it regresses a dense 3D volume aligned with the input image, then extracts the mesh as an isosurface. That makes it a natural baseline for researchers and students working on face reconstruction, 3D morphable models, or dense alignment, and the included MATLAB and Python paths let you run the pipeline without deep framework knowledge. The catch is the Torch7 and CUDA-era dependencies, so treat it as a research reference and starting point rather than something to drop into production. A Docker image and online demo help you see the method work before touching the code.