Xiao Fu (@lemonaddie0909)
2026-01-12 | โค๏ธ 369 | ๐ 54
Video generation, but 4D, dynamic, scene-consistent, and very long at the same time?!
Introducing ๐๐ฅ๐๐ง๐จ๐ฉ๐ญ๐ข๐๐๐ซ๐๐๐ฆ๐๐ซ, ๐ฆ๐ฎ๐ฅ๐ญ๐ข-๐ฏ๐ข๐๐ฐ ๐ฏ๐ข๐๐๐จ ๐ ๐๐ง๐๐ซ๐๐ญ๐ข๐จ๐ง ๐ฐ๐ข๐ญ๐ก ๐ฅ๐จ๐ง๐ -๐ญ๐๐ซ๐ฆ ๐ฌ๐ฉ๐๐ญ๐ข๐จ-๐ญ๐๐ฆ๐ฉ๐จ๐ซ๐๐ฅ ๐ฆ๐๐ฆ๐จ๐ซ๐ฒ! The scaling secret is very simple: an autoregressive paradigm with minimal 3D inductive bias, aided with a spatially grounded memory retrieval mechanism.
๐ Project page: https://research.nvidia.com/labs/dir/plenopticdreamer/ ๐ Paper: https://arxiv.org/pdf/2601.05239
๐ ์๋ณธ ๋งํฌ
๋ฏธ๋์ด
๐ Related
Auto-generated - needs manual review