Abstract
We present Kaolin, a PyTorch library aiming to accelerate 3D deep learningresearch. Kaolin provides efficient implementations of differentiable 3Dmodules for use in deep learning systems. With functionality to load andpreprocess several popular 3D datasets, and native functions to manipulatemeshes, pointclouds, signed distance functions, and voxel grids, Kaolinmitigates the need to write wasteful boilerplate code. Kaolin packages togetherseveral differentiable graphics modules including rendering, lighting, shading,and view warping. Kaolin also supports an array of loss functions andevaluation metrics for seamless evaluation and provides visualizationfunctionality to render the 3D results. Importantly, we curate a comprehensivemodel zoo comprising many state-of-the-art 3D deep learning architectures, toserve as a starting point for future research endeavours. Kaolin is availableas open-source software at https://github.com/NVIDIAGameWorks/kaolin/.