Hacker News: priowisehttps://news.ycombinator.com/user?id=priowiseHacker News RSShttps://hnrss.org/hnrss v2.1.1Tue, 21 Apr 2026 16:44:22 +0000<![CDATA[New comment by priowise in "LoGeR – 3D reconstruction from extremely long videos (DeepMind, UC Berkeley)"]]>Makes sense, thanks for the explanation. The compressed map representation + test-time training part sounds especially interesting.

Does the approach hold up well when the environment changes over time (lighting, objects moved, etc.), or does it assume mostly static scenes?

]]>Wed, 11 Mar 2026 14:29:42 +0000https://news.ycombinator.com/item?id=47336033priowisehttps://news.ycombinator.com/item?id=47336033https://news.ycombinator.com/item?id=47336033