GeoTexDensifier: Geometry-Texture-Aware Densification for High-Quality Photorealistic 3D Gaussian Splatting

Hanqing Jiang, Xiaojun Xiang, Han Sun, Hongjie Li, Liyang Zhou, Xiaoyu Zhang, Guofeng Zhang

Abstract

3D Gaussian Splatting (3DGS) has recently attracted wide attention in various areas such as 3D navigation, Virtual Reality (VR), and 3D simulation, due to its photorealistic and efficient rendering performance. High-quality reconstruction of 3DGS relies on sufficient splats and a reasonable distribution of these splats to fit real geometric surface and texture details, which turns out to be a challenging problem. We present GeoTexDensifier, a novel geometry-texture-aware densification strategy to reconstruct high-quality Gaussian splats which better comply with the geometric structure and texture richness of the scene. Specifically, our GeoTexDensifier framework carries out an auxiliary texture-aware densification method to produce a denser distribution of splats in fully textured areas, while keeping sparsity in low-texture regions to maintain the quality of Gaussian point cloud. Meanwhile, a geometry-aware splitting strategy takes depth and normal priors to guide the splitting sampling and filter out the noisy splats whose initial positions are far from the actual geometric surfaces they aim to fit, under a Validation of Depth Ratio Change checking. With the help of relative monocular depth prior, such geometry-aware validation can effectively reduce the influence of scattered Gaussians to the final rendering quality, especially in regions with weak textures or without sufficient training views. The texture-aware densification and geometry-aware splitting strategies are fully combined to obtain a set of high-quality Gaussian splats. We experiment our GeoTexDensifier framework on various datasets and compare our Novel View Synthesis results to other state-of-the-art 3DGS approaches, with detailed quantitative and qualitative evaluations to demonstrate the effectiveness of our method in producing more photorealistic 3DGS models.

Introduction

3D Gaussian Splatting (3DGS) has revolutionized real-time 3D reconstruction, but it faces a significant "splat distribution" problem. To capture a scene, 3DGS starts with a sparse point cloud and "densifies" it by splitting or cloning Gaussians to fill in details. Current methods typically rely on simple positional gradients to decide where to add more splats. This often leads to under-reconstruction, where fine details in highly textured areas (like intricate carpet patterns) are missed, or over-reconstruction, which creates a noisy mess of floating splats in low-texture areas (like flat white walls). These "floaters" waste memory and create visual artifacts.

Methodology

The GeoTexDensifier framework introduces two main strategies to optimize splat distribution:

1. Texture-Aware Densification

Instead of relying solely on positional gradients, the system analyzes the texture richness of training images. It uses a hyperbolic tangent function activated by image gradients to weight the importance of a region. In high-texture areas, the system encourages more splitting to capture fine details. Conversely, in low-texture regions, it maintains sparsity, preventing the creation of unnecessary floaters on flat surfaces.

Target regions for densification and geometric validation in a church scene.

2. Geometry-Aware Splitting

To ensure splats are placed accurately in 3D space, the framework utilizes monocular depth and normal priors (e.g., from ZoeDepth):

  • Validation of Depth Ratio Change (VDRC): When a Gaussian splits, the system checks the new splats against a relative depth map. If a new splat's depth ratio deviates significantly from its parent, it is discarded as noise.
  • Normal-Guided Splitting: Rather than random placement, the system uses surface normals to nudge new splats onto the estimated geometric surface, which is particularly effective in weakly textured areas.

Results and Evaluation

GeoTexDensifier was evaluated on standard benchmarks like Mip-NeRF 360 and Tanks and Temples, as well as custom outdoor drone captures. The method consistently outperformed state-of-the-art (SOTA) approaches such as Mini-Splatting and GeoGaussian across SSIM, PSNR, and LPIPS metrics.

GeoTexDensifier performance on a complex interior scene, showing high-fidelity texture reconstruction and quantitative metrics.

In the "Bonsai" test case, GeoTexDensifier achieved a higher frame rate (177 FPS vs. 132 FPS for Mini-Splatting) and used significantly less GPU memory (3.75GB vs. 6.61GB). Visual results show that renders of complex textures, such as carpets and architectural ornaments, are noticeably sharper and free of the ghosting artifacts common in original 3DGS.

Conclusion

GeoTexDensifier demonstrates that making the densification process smarter by incorporating texture and geometric priors leads to higher-quality 3DGS models with better memory efficiency. While the use of depth and normal priors adds some training overhead, the resulting compact and accurate representation is superior for real-time applications. Future work will explore urban-scale reconstructions and the use of generative models to fill in uncaptured regions.

Summary updated

Abstract

3D Gaussian Splatting (3DGS) has recently attracted wide attentions in various areas such as 3D navigation, Virtual Reality (VR) and 3D simulation, due to its photorealistic and efficient rendering performance. High-quality reconstrution of 3DGS relies on sufficient splats and a reasonable distribution of these splats to fit real geometric surface and texture details, which turns out to be a challenging problem. We present GeoTexDensifier, a novel geometry-texture-aware densification strategy to reconstruct high-quality Gaussian splats which better comply with the geometric structure and texture richness of the scene. Specifically, our GeoTexDensifier framework carries out an auxiliary texture-aware densification method to produce a denser distribution of splats in fully textured areas, while keeping sparsity in low-texture regions to maintain the quality of Gaussian point cloud. Meanwhile, a geometry-aware splitting strategy takes depth and normal priors to guide the splitting sampling and filter out the noisy splats whose initial positions are far from the actual geometric surfaces they aim to fit, under a Validation of Depth Ratio Change checking. With the help of relative monocular depth prior, such geometry-aware validation can effectively reduce the influence of scattered Gaussians to the final rendering quality, especially in regions with weak textures or without sufficient training views. The texture-aware densification and geometry-aware splitting strategies are fully combined to obtain a set of high-quality Gaussian splats. We experiment our GeoTexDensifier framework on various datasets and compare our Novel View Synthesis results to other state-of-the-art 3DGS approaches, with detailed quantitative and qualitative evaluations to demonstrate the effectiveness of our method in producing more photorealistic 3DGS models.
BibTeX
@article{jiang2026geotexdensifier,
  author = {Hanqing Jiang and Xiaojun Xiang and Han Sun and Hongjie Li and Liyang Zhou and Xiaoyu Zhang and Guofeng Zhang},
  title = {{GeoTexDensifier: Geometry-Texture-Aware Densification for High-Quality Photorealistic 3D Gaussian Splatting}},
  journal = {IEEE Trans. Vis. Comput. Graph.},
  year = {2026},
  doi = {10.1109/tvcg.2025.3644697},
}
Figure