Figure 8.
A four-panel illustration of the training pipeline with X-ray Rasterizer, Density Voxelizer, and Adaptive Control for radiative Gaussians reconstruction.The illustrations contain four parts. In Training Pipeline, an F D K Volume initialises Radiative Gaussians. Operation flow and gradient flow connect Adaptive Control, X-ray Rasterizer, and Density Voxelizer. The X ray Rasterizer renders images compared with Real images using L 1 and D S S I M losses. The Density Voxelizer forms a Tiny Volume with T V Regularization. In X-ray Rasterizer, World Space transforms to Ray Space, performs projection to 2 D Gaussians, and composition to produce Rendered output. In Density Voxelizer, Radiative Gaussians are organised into 3 D Tile, Voxel, and Density Volume. In Adaptive Control, operations include Split, Prune, and Clone.

R2−Gaussian (Zha et al., 2024)

This Feature Is Available To Subscribers Only

or Create an Account

Close subscription notice
Close access options