SplatAnything proposes Gaussian Spacing Normalization for 3DGS
Original title: SplatAnything: Gaussian Spacing Normalization for Feed-forward 3DGS across Arbitrary Scenes
GSNorm canonicalizes Gaussian spacing in feed-forward 3DGS for consistent supervision across scenes.
Existing feed-forward 3DGS methods normalize Gaussian attributes radially, causing systematic scale shifts. SplatAnything introduces GSNorm to canonicalize robust scene-level spacing statistics. Experiments show state-of-the-art performance on diverse datasets with varying sparsity and viewpoints.
- Addresses mismatch in Gaussian-scale targets
- Provides consistent spatial reference for prediction
- Strong gains under scene scale variations
- Benchmark spans datasets, sparsity, and viewpoints
From the original article
1 min read at huggingface.co
We identify a key limitation of existing feed-forward 3DGS methods: Gaussian attributes are typically predicted in a radial-distance-normalized space that canonicalizes a global radial statistic of the reconstructed geometry, while the spacing between neighboring Gaussian centers can still vary substantially across scene scales and camera configurations.
By Hao Yu, Zhenwei Wang, Yongzhen Hu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.