---
title: Shadow-Aware 3D Gaussian Splatting (SA-3DGS)
url: https://www.emergentmind.com/topics/sa-3dgs
type: topic
---

# Shadow-Aware 3D Gaussian Splatting (SA-3DGS)

SA-3DGS, in the sense used by "ShadowGS: Shadow-Aware 3D Gaussian Splatting for Satellite Imagery," denotes a shadow-aware extension of 3D Gaussian Splatting for multi-temporal satellite imagery that explicitly models geometry-consistent cast shadows, disentangles illumination from appearance, and preserves fast rasterization-based rendering [2601.00939]. The method is motivated by the fact that multi-temporal satellite images of the same area are acquired at different dates, with varying solar elevation and azimuth, atmospheric conditions, and sensor or view geometry; shadows therefore move, stretch, and change intensity across time, so naive fusion confounds geometry and appearance and degrades reconstruction and synthesis [2601.00939]. ShadowGS addresses this with per-Gaussian albedo and near-surface reflection features, a globally shared skylight model, efficient ray marching for geometry-aware solar visibility, a remote-sensing rendering equation, a shadow consistency constraint, and a shadow map prior for sparse-view settings [2601.00939].

## 1. Terminology and scope

In the cited literature, the abbreviation "SA-3DGS" is not unique. In "ShadowGS: Shadow-Aware 3D Gaussian Splatting for Satellite Imagery," the term refers specifically to ShadowGS [2601.00939]. Other papers use the same or closely related abbreviation for distinct methods, including Scale-Adaptive Gaussian Splatting [2403.19615], Semantic-Aware Gaussian Splatting [2405.16923], a Self-Adaptive Compression method for 3DGS [2508.03017], and a sparse-aware flat-minima optimization framework for 3DGS [2607.00885].

| Expansion | Paper | Focus |
|---|---|---|
| Shadow-Aware 3D Gaussian Splatting | "ShadowGS: Shadow-Aware 3D Gaussian Splatting for Satellite Imagery" [2601.00939] | Multi-temporal satellite imagery, shadow disentanglement, geometry-consistent shadows |
| Scale-Adaptive Gaussian Splatting | "SA-GS: Scale-Adaptive Gaussian Splatting for Training-Free Anti-Aliasing" [2403.19615] | Test-time anti-aliasing |
| Semantic-Aware Gaussian Splatting | "SA-GS: Semantic-Aware Gaussian Splatting for Large Scene Reconstruction with Geometry Constrain" [2405.16923] | Semantic guidance and geometry regularization |
| Self-Adaptive Compression | "SA-3DGS: A Self-Adaptive Compression Method for 3D Gaussian Splatting" [2508.03017] | Pruning, codebooks, repair |
| Sparse-aware 3DGS via flat minima | "Improving Sparse-View 3DGS Generalization via Flat Minima Optimization" [2607.00885] | Sparse-view generalization |

This ambiguity is consequential because the ShadowGS formulation is tied to remote-sensing physics rather than to anti-aliasing, semantic regularization, compression, or sparse-view flat-minima optimization. A plausible implication is that acronym-based citation or implementation lookup can be error-prone unless the paper title or arXiv identifier is specified explicitly.

## 2. Problem setting, inputs, and reconstruction pipeline

ShadowGS is formulated for multi-temporal satellite images with known RPC camera models, acquisition timestamps, and solar angles [2601.00939]. It handles RGB imagery on DFC2019 and pansharpened multispectral imagery on IARPA2016; the ground sampling distance is 0.3 m/pixel [2601.00939]. The method refines RPCs via bundle adjustment to obtain poses and a sparse point cloud for Gaussian initialization, then approximates the refined RPCs with a local pinhole camera in the local tangent plane; the average reprojection error for RPC fitting is less than 0.5 px [2601.00939]. Orthorectified DEMs are not required, and DSMs are used for evaluation only [2601.00939].

The multi-temporal registration stage co-registers images via BA and SfM using the refined RPCs [2601.00939]. For each timestamp $t$, the solar direction $\mathbf{s}_t$ is taken from metadata and used to compute ray-marched solar visibility [2601.00939]. The training and inference flow is: initialize 3D Gaussians from SfM points; render via 3DGS rasterization; compute solar visibility by ray marching; apply the remote-sensing rendering equation to compose color from albedo and illumination; optimize photometric, geometry, and shadow-related losses; then render novel views under arbitrary sun directions using the disentangled model at inference [2601.00939].

This pipeline is designed for scenes in which the same geometry is observed under different illumination regimes. The central methodological claim is that shadow behavior should be treated as a structured, geometry-linked phenomenon rather than as unmodeled appearance noise [2601.00939].

## 3. Gaussian representation, projection, and geometric quantities

Each 3D Gaussian is parameterized by center $\mu \in \mathbb{R}^3$, scale $s \in \mathbb{R}^3$, rotation quaternion $q \in \mathbb{R}^4$, opacity $o$, and appearance features represented with spherical harmonics [2601.00939]. Its covariance is

$$
\Sigma = R S S^{\top} R^{\top},
$$

where $R$ is the rotation from $q$ and $S$ is a diagonal scaling matrix derived from $s$ [2601.00939]. The Gaussian itself is

$$
G(\mathbf{x}) = \exp\!\left(-(\mathbf{x}-\boldsymbol{\mu})^{\top}\Sigma^{-1}(\mathbf{x}-\boldsymbol{\mu})\right).
$$

Projection uses EWA splatting, with projected 2D covariance

$$
\Sigma' = J\,W\,\Sigma\,W^{\top}J^{\top},
$$

where $W$ maps world-to-camera coordinates and $J$ is the Jacobian of the projection [2601.00939]. Rendering follows standard front-to-back alpha compositing,

$$
C = \sum_{i} T_i\,\alpha_i\,c_i,\quad T_i = \prod_{j=1}^{i-1}(1-\alpha_j),
$$

with $c_i$ the color of the $i$-th splat [2601.00939].

ShadowGS also derives depth and normal quantities from ray–Gaussian intersections. For a pixel $(u,v)$, the intersection of the camera ray with a 3D Gaussian induces a 1D Gaussian along the ray, and the peak defines the ray–Gaussian intersection depth contribution $d_i$ [2601.00939]. Depth and normals are alpha-blended, and the method applies a depth–normal consistency loss,

$$
\mathcal{L}_n = 1 - \mathbf{N}^{\top}\tilde{\mathbf{N}},
$$

where $\tilde{\mathbf{N}}$ is the normal estimated from the rendered depth map via finite differences [2601.00939]. This regularization couples geometric smoothness with the splatted representation and supports more stable shadow reasoning.

## 4. Physics-based remote-sensing rendering and disentanglement

A central component of ShadowGS is the remote-sensing rendering equation. The sun direction is computed from metadata as

$$
\mathbf{s}(\phi,\theta) =
\begin{bmatrix}
\cos\theta\cos\phi\\
\cos\theta\sin\phi\\
\sin\theta
\end{bmatrix},
$$

where $\theta$ is elevation above the horizon and $\phi$ is azimuth from the $x$-axis [2601.00939].

The method disentangles four components [2601.00939]:

- Albedo $\rho$ or $F$, parameterized by per-Gaussian spherical harmonics.
- Near-surface reflection $L_n$, also parameterized by per-Gaussian spherical harmonics and intended to capture low-order interreflections.
- Skylight $L_{\text{sky}}$, parameterized by a globally shared low-order spherical harmonic model.
- Direct sunlight $L_{\text{sun}}$ and solar visibility $S_{\text{sun}}$, where the sun is treated as a directional source and occlusion yields $S_{\text{sun}} \in [0,1]$.

After rasterization, per-pixel albedo and illumination maps are composited [2601.00939]. The direct and indirect terms are then written as

$$
L_{\text{dir}} = L_{\text{sun}}\;\max\!\bigl(0,\mathbf{n}\cdot\mathbf{s}\bigr),
$$

$$
L_{\text{ind}} = L_{\text{sky}} + L_n,
$$

$$
L_{\text{total}} = S\cdot L_{\text{dir}} + (1-S)\cdot L_{\text{ind}},
$$

$$
C = F \cdot L_{\text{total}}.
$$

The paper also gives a simplified formulation in which the direct term is normalized and $S$ gates direct versus indirect illumination:

$$
L_{\text{total}} = S + (1-S)\,(L_{\text{sky}}+L_n),\qquad C = F\cdot L_{\text{total}}.
$$

No explicit atmospheric scattering terms are modeled; instead, multi-temporal irradiance differences are absorbed by learned spherical harmonic coefficients and, optionally, per-image $L_{\text{sun}}$ scaling [2601.00939]. The resulting separation is intended to stabilize albedo across timestamps while allowing illumination to vary with sun geometry, thereby reducing geometry–radiance ambiguity [2601.00939]. This suggests that the method treats temporal illumination variation as a first-class latent factor rather than as residual appearance noise.

## 5. Geometry-consistent cast shadows and solar visibility

ShadowGS computes geometry-consistent cast shadows with efficient ray marching [2601.00939]. Its acceleration structure is a stretched icosahedron BVH over Gaussians, following 3D Gaussian Ray Tracing [2601.00939]. The per-Gaussian bounding volume is scaled with opacity and geometry as

$$
\mathbf{a} \leftarrow \boldsymbol{\mu} + \left(\sqrt{2\log\frac{o}{o_{\min}}}\right)\,SR^{\top}\mathbf{a},
$$

with transparency threshold $o_{\min}=10^{-3}$ ensuring that the BVH covers the effective Gaussian support [2601.00939].

For each Gaussian center $\boldsymbol{\mu}_0$ and sun ray direction $\mathbf{r}=\mathbf{s}$, the method marches with a fixed step size through the BVH; for each intersected Gaussian $i$, it computes the 1D intersection peak along the ray,

$$
\tau = \frac{(\boldsymbol{\mu}_0-\boldsymbol{\mu}_i)^{\top}\Sigma_i^{-1}\mathbf{r}}{\mathbf{r}^{\top}\Sigma_i^{-1}\mathbf{r}},
$$

and the response along the ray,

$$
\tilde{\alpha}_i = o_i\, G^{1D}_i(\tau).
$$

Solar visibility for the current Gaussian is then aggregated multiplicatively:

$$
S_{\text{sun}} = \prod_{i=1}^{k}\bigl(1-\tilde{\alpha}_i\bigr),
$$

where $k$ is the number of intersections along the ray [2601.00939]. The paper characterizes this as producing soft, geometry-aware cast shadows without aliasing, with integration into splatting through the aggregated $S$ map [2601.00939]. Shadow masks are rendered per view and per sun direction by rasterizing $S_{\text{sun}}$ to the image plane and alpha blending it analogously to the other composited maps [2601.00939].

A key physical observation exploited by ShadowGS is that, for satellite imaging, light rays from the sun and viewing rays are approximately parallel [2601.00939]. When view and sun directions are collinear, cast shadows are self-occluded and should not be visible in the image [2601.00939]. The method enforces this with a shadow consistency constraint: it renders a virtual shadow map $S_v$ with view direction aligned to the sun and penalizes deviation from full illumination,

$$
\mathcal{L}_{S_1} = \left\| S_v - \mathbf{1} \right\|_1.
$$

Two practical configurations are reported: a fixed camera viewpoint with sun aligned to view, and a setting in which both sun and camera directions are adjusted to be perpendicular to the scene surface [2601.00939]. The stated effect is to encourage splats to align to true surfaces and to increase opacity, improving geometric accuracy and novel-view synthesis [2601.00939].

## 6. Sparse-view regularization, outputs, and reported performance

Sparse views cause overfitting and weak constraints in ShadowGS [2601.00939]. To stabilize optimization under limited inputs, the method introduces a shadow prior from FDRNet, described as a self-supervised shadow detector with low false-negative rate on satellite data [2601.00939]. The rendered shadow map $S$ is supervised against an FDRNet mask $\hat{S}$ using BCE,

$$
\mathcal{L}_{S_3} = -\Bigl(S\log_2(\hat{S}) + (1-S)\log_2(1-\hat{S})\Bigr).
$$

Vegetation is excluded using NDVI for multispectral imagery or DEVI for RGB imagery in order to mitigate false positives in the shadow prior [2601.00939]. The prior is therefore not treated as a universal shadow oracle, but as a constrained regularizer applied where its failure modes are known.

The resulting system renders novel views under arbitrary sun directions using the disentangled representation [2601.00939]. According to the abstract, extensive experiments demonstrate that ShadowGS outperforms current state-of-the-art methods in shadow decoupling accuracy, 3D reconstruction precision, and novel view synthesis quality, with only a few minutes of training [2601.00939]. The same abstract states that the method exhibits robust performance across various settings, including RGB, pansharpened, and sparse-view satellite inputs [2601.00939].

These claims position ShadowGS as a remote-sensing-specific 3DGS variant whose main contribution is the incorporation of shadow physics into both the forward model and the optimization objective. A plausible implication is that, in satellite reconstruction, accurate shadow handling is not merely a photometric refinement but a geometric prior.

## 7. Relation to other 3DGS research directions

Within the broader 3DGS literature represented in the data, ShadowGS occupies a distinct niche. "SA-GS: Scale-Adaptive Gaussian Splatting for Training-Free Anti-Aliasing" addresses test-time anti-aliasing by applying a 2D scale-adaptive filter per projected Gaussian and, optionally, exact per-pixel integration or super-sampling [2403.19615]. "SA-GS: Semantic-Aware Gaussian Splatting for Large Scene Reconstruction with Geometry Constrain" injects semantic masks from Grounded-SAM or DINO into Gaussian shape control, expected Gaussian counts, and probability density-based point cloud extraction [2405.16923]. "SA-3DGS: A Self-Adaptive Compression Method for 3D Gaussian Splatting" focuses on pruning, importance-aware codebook compression of spherical harmonics, and codebook repair with a residual MLP [2508.03017]. "Improving Sparse-View 3DGS Generalization via Flat Minima Optimization" treats Gaussian parameters as trainable weights and regularizes sparse-view learning through anisotropy-aware perturbations and periodic reinitialization [2607.00885]. In a different sensing regime, "SAR-GS: 3D Gaussian Splatting for Synthetic Aperture Radar Target Reconstruction" adapts Gaussian splatting to SAR geometry and attenuation via the SAR Differentiable Gaussian Splatting Rasterizer [2506.21633].

Against that background, ShadowGS is specifically characterized by four coupled elements: per-Gaussian albedo and near-surface reflection features, a globally shared skylight model, efficient ray marching for geometry-aware solar visibility, and physics-based shading that composes direct and indirect illumination [2601.00939]. Its defining emphasis is therefore neither anti-aliasing nor semantic grouping nor compression nor sparse-view flat-minima optimization, but the explicit modeling of geometrically consistent shadows in multi-temporal satellite imagery [2601.00939].

A common misconception would be to read ShadowGS as a generic illumination-aware extension applicable unchanged to arbitrary 3DGS settings. The paper’s formulation is more specific: it assumes multi-temporal satellite inputs, known RPC camera models, acquisition timestamps, and solar angles, and it exploits the approximately parallel geometry of sun and satellite rays in its shadow consistency constraint [2601.00939]. The method is thus best understood as a remote-sensing-specialized 3DGS formulation rather than as a general-purpose replacement for all SA-3DGS variants.

Source: https://www.emergentmind.com/topics/sa-3dgs