AaronJackson/vrn
Vrn
vrn ist ein Tool zur 3D-Gesichtsrekonstruktion, das aus einzelnen zweidimensionalen Bildern dreidimensionale volumetrische Repräsentationen von menschlichen Gesichtern generiert. Es nutzt ein volumetrisches Convolutional Neural Network Regressionsmodell, um 3D-Volumendaten direkt aus Bildpixeln vorherzusagen.
Das System konvertiert diese volumetrischen Vorhersagen durch Isosurface-Extraktion und Vertex-Färbung in 3D-Meshes. Zudem wendet es realistische Oberflächendetails an, indem es zweidimensionale Bildpixel mittels Nearest-Neighbor-Texturprojektion auf das resultierende 3D-Mesh mappt.
Das Projekt bietet Funktionen für die Tiefenschätzung aus Einzelbildern und die digitale Modellierung menschlicher Gesichter.
Features
- Single-Image 3D Reconstructions - Generates complete 3D volumetric representations of human faces from a single two-dimensional input image.
- Direct Voxel Prediction - Implements a volumetric CNN regression model that predicts 3D volume data directly from image pixels.
- Volumetric Regression Architectures - Implements a deep learning architecture that predicts 3D volume data directly from 2D image inputs.
- Volumetric CNN Regression - Utilizes a volumetric convolutional neural network regression model to predict 3D volume data from pixels.
- 3D Face Reconstruction - Provides a system for generating 3D volumetric representations of human faces from single 2D images.
- Volumetric Reconstruction Tools - Provides a tool for creating dense volumetric 3D face models from single images.
- Monocular Depth Estimators - Infers three-dimensional spatial depth and volume from single 2D facial photographs.
- Human Digitalization - Generates 3D meshes with realistic surface textures suitable for digital human modeling.
- Isosurface Extractors - Converts volumetric predictions into 3D meshes by calculating the isosurface and applying vertex colors.
- Image-to-3D Texture Engines - Transfers 2D image textures onto 3D digital face representations using nearest-neighbor sampling.
- Volumetric Mesh Extraction - Converts volumetric density fields into discrete polygonal triangle meshes through isosurface extraction.
- Texture Projection Mapping - Maps 2D image pixels onto the resulting 3D mesh using nearest-neighbor projection for realistic surface details.
- Texture Sampling - Projects colors from 2D images onto a 3D mesh using nearest-neighbor pixel sampling.
- Face Alignment And Reconstruction - Volumetric CNN regression for 3D reconstruction from single images.
- Generative Models - Volumetric CNN regression for 3D face reconstruction.
- Faces: - Listed in the “Faces:” section of the Curated List Of Awesome 3D Morphable Model Software And Data awesome list.
