---
title: 'StableGS: Floater-Free 3D Gaussian Splatting'
url: https://www.emergentmind.com/topics/stablegs
type: topic
---

# StableGS: Floater-Free 3D Gaussian Splatting

Searching arXiv for the StableGS paper and closely related 3D Gaussian Splatting works.
Searching for "StableGS 3D Gaussian Splatting".
StableGS is a framework for 3D Gaussian Splatting (3DGS) that targets a specific optimization pathology of novel view synthesis: floater artifacts, namely small, translucent Gaussians that drift into empty space and persist during training. Introduced in 2025, the method augments the standard 3DGS pipeline with cross-view depth consistency constraints, a dual-opacity Gaussian formulation, and DUSt3R-based depth priors with dense-reconstruction initialization, with the stated goal of eliminating floaters while preserving translucent materials and improving weakly textured geometry [2503.18458]. In the lineage of 3DGS-based radiance-field rendering [2308.04079], StableGS shifts supervision from purely photometric fitting toward a joint photometric-geometric regime.

## 1. Problem formulation and diagnosis of floater artifacts

StableGS is motivated by the observation that vanilla 3DGS often “locks in” floaters during training [2503.18458]. In the standard 3DGS rendering equation,
\[
\sigma_k(x)=\alpha_k\exp\!\Bigl(-\tfrac12(x-\mu_k)^T\Sigma_k^{-1}(x-\mu_k)\Bigr),
\]
the opacity $\alpha_k$ and color $c_k$ are jointly optimized. StableGS identifies two intertwined causes of failure.

The first is opacity-color coupling. The reported mechanism is that clusters of floaters can produce blended colors $c_f$ that approximately match the background, so $\partial L/\partial \alpha_k$ and $\partial L/\partial c_k$ vanish. When this occurs, the optimizer cannot “turn off” the spurious Gaussians, and training stalls in local minima [2503.18458]. The second is lack of geometric supervision: purely photometric losses, including $L1$ or SSIM on rendered RGB, do not distinguish true foreground structure from floaters [2503.18458].

This diagnosis is central to the method’s identity. StableGS does not treat floaters primarily as a rasterization artifact or a post hoc pruning problem; rather, it treats them as an optimization degeneracy induced by coupled appearance-geometry fitting under insufficient geometric constraints. This suggests a reinterpretation of 3DGS instability as a failure of identifiability under photometric supervision alone.

## 2. Cross-view depth consistency as floater suppression

The first core component is a cross-view depth consistency loss. For two frames $i$ and $j$, StableGS renders depth maps $\bar D_i$ and $\bar D_j$ alongside color, then uses known camera extrinsics to warp one depth map into the other view, denoted $\mathcal W_{i\to j}(\bar D_i)$ [2503.18458]. The consistency term is
\[
\mathcal L_{\rm consis}
  = \sum_{(i,j)\in\mathcal P}\sum_{(u,v)\in\mathcal V}
    \Bigl|\mathcal W_{i\to j}(\bar D_i(u,v))-\bar D_j(u,v)\Bigr|
  + \Bigl|\mathcal W_{j\to i}(\bar D_j(u,v))-\bar D_i(u,v)\Bigr|.
\]

Here $\mathcal P$ is a set of view-pairs chosen to satisfy covisibility, angular, and baseline constraints [2503.18458]. The loss is added to the photometric objective with coefficient $\lambda_{\rm consis}=0.05$, and the reported empirical effect is that it drives $\alpha_k\to 0$ for floaters [2503.18458].

The mechanism is geometric: a floating Gaussian that projects plausibly in one view but occupies inconsistent depth in another incurs a direct penalty. Because the constraint is mutual and cross-view, it attacks the spatial inconsistency that photometric fitting alone cannot detect. StableGS therefore turns multi-view agreement into an explicit regularizer on Gaussian occupancy rather than relying on RGB residuals to indirectly recover geometry.

## 3. Dual-opacity Gaussian model and translucent materials

A potential problem with aggressive depth consistency is that it could suppress legitimate translucent objects, especially glass. StableGS addresses this with a dual-opacity Gaussian model in which each Gaussian $G_k$ carries two opacities, $\alpha_k$ and an auxiliary $\alpha'_k$ [2503.18458]. The densities are defined as
\[
\sigma_k(x)=\alpha_k\,\exp\!\bigl(\cdots\bigr), \qquad
\sigma'_k(x)=\alpha_k\,\alpha'_k\,\exp\!\bigl(\cdots\bigr).
\]

Rendering in “standard” mode uses $\sigma$ to produce $(\bar C^s,\bar D^s)$, while “dual-opacity” mode uses $\sigma'$ to produce $(\bar C^o,\bar D^o)$ [2503.18458]. Over two neighboring frames $i,j$, the combined color loss is
\[
\mathcal L_{\rm color}
= \sum_{k\in\{i,j\}}
\Bigl(
\mathcal L_{\rm 3DGS}(\bar C^o_k,C_k^{\rm gt})
+ \|\bar C^s_k-C_k^{\rm gt}\|_1
\Bigr).
\]

The key relation is $\sigma' \le \sigma$. StableGS uses this asymmetry so that floaters, which produce anomalously small depths, are eliminated in $\bar D^s$ under depth consistency, while $\alpha'$ can restore true translucency in $\bar D^o$ and $\bar C^o$ [2503.18458]. A common misconception is therefore avoided: the method is not a blanket suppression of low-opacity structure. Its stated design is to decouple geometric opacity from material appearance specifically to avoid destroying translucent objects when enforcing depth consistency.

## 4. DUSt3R depth priors and dense initialization

StableGS states that weakly textured regions, such as white walls, still challenge pure splatting [2503.18458]. Its third component injects external geometric priors through DUSt3R [2312.14132]. The pipeline runs pairwise depth estimation to produce point maps $\Phi(i,j)$, then solves a global scale optimization:
\[
\min_S \sum_{(i,j)\in\mathcal P}\bigl\|w_{i,j}\odot\bigl(s_{i,j}T^i
[x_{i,j},y_{i,j}]^T - s_{j,i}T^j[x_{i,j},y_{i,j}]^T\bigr)\bigr\|.
\]
This yields consistent depth priors $d_{i,j}$ and confidences $w_{i,j}$ [2503.18458].

The corresponding depth-prior loss is
\[
\mathcal L_{\rm prior}
= \sum_{(i,j)\in\mathcal P}\frac{1}{\|d_{i,j}\|_0}
\bigl\|w_{i,j}\odot(\bar D^s_i-d_{i,j})\bigr\|_1
+ (\text{symmetrically for }j\to i),
\]
weighted by $\lambda_{\rm prior}=0.005$ [2503.18458].

StableGS further applies a learnable correction
\[
\phi(d;r,l)=d^r+l\cdot d
\]
to DUSt3R depths, fuses them via TSDF to produce an initial dense mesh, and scatters Gaussians for robust initialization [2503.18458]. This part of the framework is significant because it does not merely regularize the final geometry; it changes the starting point of optimization through dense initialization. A plausible implication is that the method reduces both poor local minima and slow geometric convergence in low-texture areas where photometric gradients vanish.

## 5. Optimization pipeline and objective

The overall optimization pipeline begins with COLMAP for coarse poses and sparse points [2503.18458]. For each pair $(i,j)\in\mathcal P$, StableGS computes DUSt3R outputs $\Phi(i,j)$, solves the global scale problem, refines depths via $\phi(\cdot)$, and fuses a TSDF to obtain the initial Gaussian set $\mathcal G$ [2503.18458].

Training then proceeds for $t=1$ to $T$, with the example value $T=30$ k iterations. At each iteration, the method renders $\bar C^s$, $\bar D^s$, $\bar C^o$, and $\bar D^o$ for minibatch frames, and optimizes
\[
\mathcal L_{\rm total}
= \mathcal L_{\rm color}
+ \lambda_{\rm consis}\,\mathcal L_{\rm consis}
+ \lambda_{\rm prior}\,\mathcal L_{\rm prior}.
\]
In the provided LaTeX form, this is written as
\[
\mathcal L_{\rm total}
= \underbrace{\sum_{k\in\{i,j\}\Bigl(\mathcal L_{\rm 3DGS}(\bar C^o_k,C_k^{\rm gt})
+ \|\bar C^s_k - C_k^{\rm gt}\|_1\Bigr)}_{\mathcal L_{\rm color}
+ 0.05\,\mathcal L_{\rm consis}
+ 0.005\,\mathcal L_{\rm prior}.
\]

Backpropagation updates the Gaussian parameters
\[
\{\alpha_k,\alpha'_k,\mu_k,q_k,s_k,c_k\}
\]
[2503.18458]. This unified objective makes the method structurally different from approaches that rely only on RGB reconstruction plus densification heuristics. StableGS explicitly combines appearance fitting, multi-view geometric consensus, and external depth priors within a single optimization loop.

## 6. Empirical behavior, interpretation, and scope

StableGS is reported to outperform existing state-of-the-art methods across standard benchmarks including MipNeRF360, Blender, and Tanks & Temples [2503.18458]. The detailed example provided is the four indoor MipNeRF360 scenes:

| Method | PSNR / SSIM / LPIPS | Setting |
|---|---:|---|
| 3DGS (baseline) | 30.46 / 0.9170 / 0.1975 | four indoor MipNeRF360 scenes |
| Mip-Splatting | 30.97 / 0.9242 / 0.1870 | four indoor MipNeRF360 scenes |
| RadeGS | 30.51 / 0.9213 / 0.1887 | four indoor MipNeRF360 scenes |
| StableGS | 31.11 / 0.9222 / 0.1892 | four indoor MipNeRF360 scenes |
| StableGS + prior | 31.19 / 0.9245 / 0.1838 | four indoor MipNeRF360 scenes |
| StableGS + prior + DenseInit | 31.12 / 0.9246 / 0.1829 | four indoor MipNeRF360 scenes |

The qualitative claims are equally specific. StableGS renders crisp fine details, with “no grass-blurring,” eliminates sky-region floaters, and accurately reproduces glass and specular materials via its dual-opacity mode [2503.18458]. In weak-texture corners such as blank walls, the depth priors anchor geometry where pure photometric gradients vanish [2503.18458].

Within the broader Gaussian-splatting literature, StableGS addresses a static-scene training instability—floater artifacts rooted in opacity-color coupling and weak geometry supervision. By contrast, dynamic-scene methods such as SharpTimeGS focus on temporal visibility, motion drift, and densification balance in 4D Gaussian representations rather than on floater suppression in static 3DGS [2602.02989]. The term “StableGS” is also used in unrelated numerical contexts, including stochastic optimization and ghost stabilisation for the Material Point Method [2209.14937; 2209.10955]. In 3D vision, however, StableGS denotes the floater-free 3D Gaussian Splatting framework centered on cross-view depth consistency, dual-opacity disentanglement, and DUSt3R-guided initialization [2503.18458].

Source: https://www.emergentmind.com/topics/stablegs