Weakly-Supervised Shape Multi-Completion of Point Clouds by Structural Decomposition.

Summary: Have you ever tried to guess what a whole puzzle looks like when half the pieces are missing? Computers struggle with this too, especially when trying to turn incomplete 3D scans (called "point clouds") into full 3D shapes. A new method helps computers do this much better by breaking objects down into a basic frame and smaller parts. Instead of needing perfect 3D examples to learn from, the computer uses flat images to fill in the missing details. It even uses a special creative tool to guess different ways the missing parts might look. This new trick makes the computer 38% better at building full 3D shapes from broken scans!