https://mllog.dev/en/tags/3d/Recent content in 3D on MLLog.devhttps://mllog.dev/images/default_mllog.pnghttps://mllog.dev/images/default_mllog.pngHugo -- 0.147.9enSat, 07 Mar 2026 00:00:00 +0000https://mllog.dev/en/posts/utonia-one-encoder-all-point-clouds/Sat, 07 Mar 2026 00:00:00 +0000https://mllog.dev/en/posts/utonia-one-encoder-all-point-clouds/<p>A LiDAR on a self-driving car, a depth camera in a home robot, a satellite scanner, and a CAD model from a 3D printer — each produces a <span class="glossary-term" tabindex="0"> <span class="glossary-word">point cloud</span> <span class="glossary-tooltip"> <strong>point cloud</strong> <span class="glossary-def">A set of 3D points (x, y, z) representing the shape of an object or scene. Each point can carry additional attributes: color, normal, intensity.</span> </span> </span> , but with radically different density, scale, and geometry. Until now, each domain required its own model. The paper <strong>&ldquo;Utonia: Toward One Encoder for All Point Clouds&rdquo;</strong> breaks this pattern — one encoder, 137M parameters, five domains, and emergent behaviors nobody expected.</p>