Anchor3DGS takes uncalibrated multi-view images and, in a single forward pass, outputs a set of 3D Gaussians for the scene.
Anchor-based 3D Gaussians cut primitives tenfold without losing quality
A new feed-forward framework replaces pixel-aligned predictions with entropy-guided 3D anchors, achieving state-of-the-art novel view synthesis while using an order of magnitude fewer Gaussians.
Chinese Tech
Sheng Ye · Jiangke Lin · Fudong Wang · Yi Yuan · Zhen-Hui Dong · Yong-Jin Liu
Tsinghua University · Ant Group
Research Digest··3 min read
The authors present Anchor3DGS, a pose-free feed-forward 3D Gaussian Splatting method that replaces the standard pixel-aligned prediction of one Gaussian per pixel with a compact anchor-based representation.
Why this paper
From Ant Group and Tsinghua University
In one line
Anchor3DGS achieves high-fidelity novel view synthesis with an order of magnitude fewer Gaussians by replacing pixel-aligned prediction with compact anchor-based representation.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§