NVIDIAGameWorks/kaolin
Kaolin
Kaolin is a PyTorch 3D deep learning library providing a comprehensive suite of tools for 3D geometry processing, physics simulation, data visualization, and gradient-based rendering for computer vision.
The library includes a differentiable 3D renderer and a geometry processing toolkit for converting and transforming 3D representations such as meshes and point clouds. It also features a 3D physics simulation engine to calculate physical interactions and collisions between three-dimensional objects and scenes.
The toolkit provides utilities for 3D data visualization, including the creation of interactive views and turntable animations. Additional capabilities cover 3D dataset management, data preprocessing, and 3D representation rendering.
Features
- Differentiable Rasterizers - Ships a differentiable rasterizer that allows gradients to flow from 2D images back into 3D spatial parameters.
- PyTorch-Based Frameworks - Acts as a PyTorch-based framework for accelerating research in 3D computer vision and deep learning.
- PyTorch Tensor Operations - Performs all 3D geometric calculations using PyTorch tensor operations for GPU acceleration and automatic differentiation.
- 3D Point Cloud Learning - Enables deep learning research on 3D geometric data including meshes and point clouds using PyTorch.
- 3D Physics Engines - Includes a 3D physics engine to calculate physical interactions and collisions between objects and scenes.
- 3D Math and Geometry Toolkits - Provides a comprehensive toolkit for converting, managing, and transforming 3D representations such as meshes and point clouds.
- Differentiable Rendering - Provides a differentiable rendering pipeline allowing 3D geometry optimization via gradients from 2D images.
- CUDA Compute Kernels - Provides custom CUDA compute kernels in C++ to parallelize high-speed 3D physics and spatial transformations.
- 3D Representation Conversions - Provides utilities for transforming geometric data between formats like meshes, point clouds, and volumetric structures.
- Coordinate-Based Spatial Mappings - Implements coordinate-based spatial mapping to represent 3D geometry as functions or grids in Cartesian space.
- Voxel-Based Grids - Organizes spatial data into voxel-based grids to optimize collision detection and volumetric operations.
- Mesh-To-Pointcloud Sampling - Provides weighted area sampling to convert surface meshes into discrete point sets for deep learning input.
- 3D Spatial Preprocessing - Implements 3D spatial preprocessing pipelines to transform data formats for improved deep learning training speed.
- Physical Interaction Simulations - Simulates physical interactions and collisions between meshes and point clouds within 3D scenes.
- 3D Model Visualizers - Offers tools for rendering and inspecting 3D models with interactive views and turntable animations.
- 3D Scene Renderers - Produces 2D images from spatial data using lighting and camera models to support gradient-based optimization.
- Data and Graph Processing - Library for 3D deep learning research.
- Point Cloud Processing - PyTorch library for accelerating 3D deep learning research.
- Developer Tools - 3D deep learning research library.
- Processing Libraries - NVIDIA library for accelerating 3D deep learning research.
- D Segmentation, Classification and Regression - Listed in the “D Segmentation, Classification and Regression” section of the The Incredible Pytorch awesome list.
Star history
How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.
