AtomGS: Atomizing Gaussian Splatting for High-Fidelity Radiance Field

Rong Liu,Rui Xu,Yue Hu,Meida Chen,Andrew Feng

2024-05-23

Abstract:3D Gaussian Splatting (3DGS) has recently advanced radiance field reconstruction by offering superior capabilities for novel view synthesis and real-time rendering speed. However, its strategy of blending optimization and adaptive density control might lead to sub-optimal results; it can sometimes yield noisy geometry and blurry artifacts due to prioritizing optimizing large Gaussians at the cost of adequately densifying smaller ones. To address this, we introduce AtomGS, consisting of Atomized Proliferation and Geometry-Guided Optimization. The Atomized Proliferation constrains ellipsoid Gaussians of various sizes into more uniform-sized Atom Gaussians. The strategy enhances the representation of areas with fine features by placing greater emphasis on densification in accordance with scene details. In addition, we proposed a Geometry-Guided Optimization approach that incorporates an Edge-Aware Normal Loss. This optimization method effectively smooths flat surfaces while preserving intricate details. Our evaluation shows that AtomGS outperforms existing state-of-the-art methods in rendering quality. Additionally, it achieves competitive accuracy in geometry reconstruction and offers a significant improvement in training speed over other SDF-based methods. More interactive demos can be found in our website (

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

This paper attempts to address several key issues present in existing 3D Gaussian Splatting (3DGS) methods when reconstructing radiance fields:

1. **Insufficient Geometric Accuracy**: Existing 3DGS methods may prioritize enlarging large Gaussian ellipsoids during optimization while neglecting the densification of small Gaussian ellipsoids, leading to a lack of geometric details. This results in noise and blurry artifacts in the reconstructed 3D geometric structure.

2. **Poor Rendering Quality**: Although 3DGS methods perform well in real-time rendering speed, there is still room for improvement in rendering quality. Especially when dealing with areas with fine features, existing methods often fail to preserve these details well.

3. **Imbalance in the Optimization Process**: 3DGS methods lack precise constraints on geometric structures during the optimization process, leading to inaccurate alignment of Gaussian ellipsoids with the actual geometric structure, thereby affecting the final rendering effect and geometric accuracy.

To address these issues, the paper proposes the AtomGS method, which introduces two key techniques: Atomized Proliferation and Geometry-Guided Optimization to improve 3DGS methods. Specifically:

- **Atomized Proliferation**: Constrains Gaussian ellipsoids of different sizes into more uniform atomic Gaussians and provides denser coverage in areas with scene details to enhance geometric accuracy.
- **Geometry-Guided Optimization**: Introduces Edge-Aware Normal Loss to retain complex details while smoothing surfaces, thereby improving the accuracy of geometric reconstruction.

With these improvements, AtomGS surpasses existing state-of-the-art methods in both rendering quality and geometric accuracy.