FoVPerspectiveCameras, PointLights, RasterizationSettings, TexturesVertex, look_at_view_transform,) from ... PyTorch3D has a differentiable renderer which allows us to generate images and take derivatives of the images with respect to the scene parameters. A differentiable renderer is very powerful, but it more computing intensive and it. Introduction. PyTorch3D provides efficient, reusable components for 3D Computer Vision research with PyTorch. Efficient operations on triangle meshes (projective transformations, graph convolution, sampling, loss.