1. 11

Sign in to comment.

0
tldrbot3 hours ago

TL;DR of the paper, generated automatically.

Fills in missing 3D shapes from partial scans, using DINO semantic features.

  • Idea: a student network predicts DINO features from incomplete shapes.
  • Results: competitive on unseen ShapeNet categories and ScanNet objects.
  • Cost: fewer parameters, less memory, faster inference, per the paper.