Rasterization Pipeline

Rasterization pipelines are fundamental to computer graphics, aiming to efficiently render 3D scenes into 2D images. Current research focuses on improving the speed and quality of rendering using novel scene representations, such as Gaussian splatting, which represents scenes as collections of 3D Gaussians, and adapting neural radiance fields (NeRFs) for compatibility with traditional rasterization hardware. These advancements are improving the efficiency and realism of rendering complex scenes, particularly for applications like high-fidelity human performance capture and real-time rendering on mobile devices. Optimizations targeting the pipeline's atomic operations during differentiable rendering are also a key area of focus, accelerating the training process for learned 3D models.