We propose a feed-forward framework for learning compact 3D representations from unposed images. Our approach estimates only 2K Gaussians that allocated in meaningful regions to enable generalizable ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results