Zeqi Xiao (@zeqi_xiao)
2025-12-02 | โค๏ธ 38 | ๐ 3
How far can video generative models go in visuospatial intelligence? ๐ค
We propose Video4Spatial, showing that with video-only context, models can:
๐บ๏ธ Plan in 3D and ground objects ๐ฅ Follow camera-pose instructions ๐งฑ Maintain strong spatial consistency https://x.com/zeqi_xiao/status/1995992142142161040/video/1
๐ ์๋ณธ ๋งํฌ
๋ฏธ๋์ด
