Skip to main content
. 2021 Oct 28;1:740342. doi: 10.3389/fbinf.2021.740342

FIGURE 1.

FIGURE 1

An overview of our network. Batches of images of size 128 × 128 pixels are fed to the convolution layers, which reduce the batch down to 6 parameters. These are passed to the rendering pipeline along with the 3D reconstruction matrix to produce a batch of output images. The rendering parameters Rx, Ry and Rz represent the rotation in “Angle-Axis” form. Tx and Ty represent translation in the X, Y plane. S represents the predicted output-sigma. The 3D reconstruction matrix contains a list of vertices representing the predicted point-cloud.