---
title: 'Planar Gaussians: Rendering, Mapping, and Analysis'
url: https://www.emergentmind.com/topics/planar-gaussians
type: topic
---

# Planar Gaussians: Rendering, Mapping, and Analysis

Planar Gaussians denote several closely related constructions rather than a single universal primitive. In contemporary computer vision, the term most often refers to anisotropic 3D Gaussian splats whose smallest principal scale is explicitly compressed so that each primitive approximates a local tangent plane, making rendered normals and depth geometrically useful for reconstruction, meshing, articulation, and consistency losses [2511.17092]. In other settings, a planar Gaussian is literally a 2D Gaussian defined on a plane, as in bird’s-eye-view mapping, or a standard Gaussian whose center coordinates are constrained to a shared plane representation [2605.09619]. In mathematics, the phrase also appears in work on Gaussian densities on the plane, planar Gaussian fields, and large-\(N\) “planar” asymptotics of Gaussian partition functions [1604.00592].

## 1. Terminological scope and common geometric idea

Across the literature surveyed here, “planar Gaussian” has at least four technically distinct meanings. This suggests that the term is best understood by the geometric role it plays, not by a single canonical formula.

| Usage | Core object | Operational meaning |
|---|---|---|
| Planarized 3D Gaussian splat | Standard anisotropic 3D Gaussian | One principal scale is pushed toward zero, so the primitive behaves like a local surface patch |
| Literal planar Gaussian | 2D Gaussian on a plane | Density kernel on \((x,y)\), usually with anisotropic covariance and in-plane rotation |
| Plane-structured Gaussian | Standard Gaussian with constrained center | The Gaussian center is parameterized to remain on a learned or detected plane |
| Planar Gaussian in mathematical analysis | Gaussian density or field on the plane | The adjective “planar” refers to the ambient domain, not necessarily to a flattened primitive |

In the splatting literature, the common idea is to replace appearance-dominated volumetric ellipsoids by primitives that carry more explicit local surface structure. SPAGS states this directly: ordinary 3D Gaussian Splatting is poorly suited to sparse-view articulated object reconstruction because unconstrained 3D Gaussian primitives are “disorderly and irregular,” which makes reliable depth and normal estimation difficult; the remedy is to compress 3D Gaussians into planar Gaussians and optimize them with geometry-oriented regularization [2511.17092]. REArtGS++ and GigaGS adopt essentially the same mechanism: a Gaussian remains a standard oriented anisotropic 3D primitive, but its minimum scale is explicitly penalized so that the primitive becomes a thin sheet whose shortest axis serves as a normal direction [2511.17059]. 

Other works use the adjective more literally. GSMap defines a planar Gaussian as a 2D primitive on the ground plane / BEV plane with center \(\mu\in\mathbb{R}^2\), anisotropic scale \(\sigma\in\mathbb{R}^2\), and rotation \(\theta\in\mathbb{R}\), rendered as an oriented elliptical support region [2605.09619]. GSPlane, by contrast, does not alter Gaussian covariance at all; its “planar Gaussian” is a Gaussian whose center is re-parameterized as an affine combination of three basis points on a plane, so off-plane drift becomes impossible while the primitive remains in the planar class [2510.17095].

## 2. Surface-oriented planar Gaussians in 3D Gaussian Splatting

The most influential vision formulation treats a planar Gaussian as a constrained anisotropic 3D Gaussian. In SPAGS, the primitive \(\mathcal G_i\) retains the usual Gaussian center \(\mathbf p\in\mathbb R^3\), opacity \(\alpha_i\), and principal scales \(S_1,S_2,S_3\), but planarity is enforced by the loss
\[
\mathcal{L}_{\text{scale}} = \frac{1}{N_{\mathcal G}}\sum_i^{N_{\mathcal G}} \left\| \min(S_1,S_2,S_3)\right\|.
\]
Operationally, this drives the smallest principal scale toward zero, so the covariance becomes effectively rank-deficient or nearly rank-2 in practice. SPAGS further states that “Based on the planar Gaussian primitives, we can easily acquire the normal via the shortest axis and viewing direction,” and uses alpha-blended normal rendering \(N\) together with an “unbiased depth” \(\mathbf D\) following PGSR [2511.17092].

This flattened-splat interpretation is now used as a geometry backbone rather than merely a rendering convenience. In SPAGS, planar Gaussians sit between sparse-view selection and part segmentation: optimal sparse views are selected by a Gaussian information field, a structured initialization is produced and registered, Gaussians are converted to planar Gaussians, and then optimized in a coarse-to-fine schedule with color, depth, depth smoothness, few-shot diffusion, and view consistency; only afterward are part probabilities attached to each Gaussian via mask back-projection [2511.17092]. The geometric importance is borne out by ablation: replacing planar Gaussians with the depth rendering pipeline from GaussianObject worsens whole-object CD from \(9.60\) to \(30.98\), dynamic-part CD-d from \(25.87\) to \(61.52\), and static-part CD-s from \(31.02\) to \(76.25\) on PartNet-Mobility [2511.17092].

REArtGS++ extends the same planarization into articulated motion. Its scale loss
\[
\mathcal{L}_{\text{scale}} = \frac{1}{N_{\mathcal G}}\sum_i^{N_{\mathcal G}}\left\|\min(s_1,s_2,s_3)\right\|
\]
again collapses one axis, but the resulting normals and unbiased depth are then used in a temporal geometry constraint defined over the entire articulated motion interval via first-order Taylor expansion [2511.17059]. The practical effect is large: in the ablation “w/o planar GS” versus “w/ all,” Axis Ang improves from \(5.18\) to \(0.41\), Axis Pos from \(21.38\) to \(1.18\), Part Motion from \(6.23\) to \(0.22\), CD-s from \(5.84\) to \(1.41\), CD-m from \(7.57\) to \(2.38\), and CD-w from \(6.45\) to \(2.13\) [2511.17059].

GigaGS generalizes the same planar bias to large aerial scenes. It defines a flatten loss
\[
\mathcal{L}_{flatten} = \frac{1}{|\mathbb M|}\sum_{i\in\mathbb M}\big|\min(\mathbf s_i)\big|
\]
and uses the shortest axis as the Gaussian normal \(\mathbf n_i\). From these planarized splats it renders normals and plane distances, then reconstructs depth by ray-plane intersection. The paper’s own ablation is explicit that planarity alone is insufficient: flatten regularization without stronger local and multiview constraints can reduce expressiveness, while the combination of flattening, local depth-normal consistency, and multi-view planar consistency improves surface quality [2409.06685].

## 3. Structured plane-aware variants and representational alternatives

A second line of work does not necessarily flatten covariance, but instead injects plane structure into coordinates, semantics, or optimization. GSPlane is the clearest example. For each planar cluster, it estimates a plane, selects three non-collinear basis points \(F_1,F_2,F_3\), and re-parameterizes every planar Gaussian center as
\[
V'_P=\omega_1F_1+\omega_2F_2+\omega_3F_3,\qquad \omega_1+\omega_2+\omega_3=1.
\]
Planarity is therefore a constraint on center coordinates rather than on covariance eigenvalues or rotations. The method further introduces a Dynamic Gaussian Re-classifier: it selects the top 5% of planar Gaussians ranked by average gradient magnitude and reclassifies those whose gradient exceeds the average magnitude of the top 20% non-planar Gaussians back to unconstrained \(xyz\) form [2510.17095]. The paper reports that the structured planar representation improves planar metrics while reducing planar vertex counts drastically; for example, on ScanNetV2, 2DGS baseline F-score \(0.583\) becomes \(0.689\) with “Ours-full,” and planar vertex count drops from \(1.73\)M to \(946.1\)K [2510.17095].

PlanarGS keeps standard 3DGS primitives but combines language-prompted planar masks, depth priors, normal priors, Gaussian flattening, and a co-planarity loss. Its flattening term is again the minimum-scale penalty
\[
L_s=\|\min(s_1,s_2,s_3)\|_1,
\]
but the more distinctive term is the plane-fit supervision: for each planar region \(p_m\), a plane \(\mathbf A_m^T\mathbf P=1\) is fit from rendered 3D points, and rendered depth \(\hat D\) is penalized against the induced plane depth \(D_p\) via
\[
L_p=\frac{1}{N_p}\sum_{\mathbf p\in p}\|D_p(\mathbf p)-\hat D(\mathbf p)\|_1.
\]
The total loss is
\[
L_{total}= L_{RGB}+L_s+\lambda_1 L_{dn}+\lambda_2 L_p+\lambda_3 L_{rd}+\lambda_4 L_{rn},
\]
with \(\lambda_1=0.05\), \(\lambda_2=0.5\), \(\lambda_3=0.05\), and \(\lambda_4=0.2\) [2510.23930]. On MuSHRoom, PlanarGS reports Acc \(3.95\), Comp \(5.02\), CD \(4.49\), F1 \(77.14\), and NC \(83.35\), and the ablation “w/o both co-planarity and geometric priors” degrades to CD \(6.59\) and F1 \(64.11\) [2510.23930].

AtlasGS is plane-aware in yet another sense. It builds on 2DGS-style surfel Gaussians with attributes \(\alpha\in\mathbb R\), \(s\in\mathbb R^2\), \(q\in\mathbb R^4\), \(c\in\mathbb R^3\), and semantic logits \(z\in\mathbb R^4\), then regularizes them using Atlanta-world plane indicators for floor, ceiling, and gravity-aligned wall structure [2510.25129]. The explicit 3D regularizer
\[
\mathcal L_{3D} = \sum_{i\in M_\parallel}p_\parallel(1-|\mathbf n_g^\top\mathbf n_i|)+\sum_{i\in M_\perp}p_\perp|\mathbf n_g^\top\mathbf n_i|+\sum_{i\in M_c}p_c|d_c-\mathbf n_g^\top\mathbf p_i|+\sum_{i\in M_f}p_f|d_f+\mathbf n_g^\top\mathbf p_i|
\]
is complemented by a 2D local surface term from rendered depth normals. On ScanNet, the ablation “w/o \(\mathcal L_{reg}\)” yields CD \(4.10\) and F-score \(74.23\), whereas the full model reports CD \(3.77\) and F-score \(77.98\) [2510.25129].

## 4. Dynamic, articulated, and optical extensions

Planar Gaussians are especially useful when geometry must remain coherent under articulation or under layered image formation. In articulated reconstruction, SPAGS associates each Gaussian with a part probability
\[
m_i^o=\mathcal M^o(\rho)\,\alpha_i\prod_{j=1}^{i-1}(1-\alpha_j),
\]
so part labels are back-projected onto surface-like primitives rather than amorphous volumetric blobs [2511.17092]. This makes planar Gaussians not just a meshing device but also the substrate for part-level reasoning.

REArtGS++ adds a time-continuous constraint that depends directly on the planar interpretation. It renders planar-Gaussian normals \(\mathbf N(\omega,t)\) and unbiased depth \(\mathbf D(\omega,t)\), estimates a depth-derived normal \(\bar{\mathbf N}(\omega,t)\), and aligns both the normals and their temporal derivatives:
\[
\mathcal L_{geo}=(1-\nabla \mathbf I(t_0))\left(\|\bar{\mathbf N}(\omega,t_0)-\mathbf N(\omega,t_0)\|+\|\nabla\bar{\mathbf N}(\omega,t_0)-\nabla\mathbf N(\omega,t_0)\|\right).
\]
The canonical state \(t^*=0.5\) is chosen because no motion is applied there and \(N(\omega,t^*)\) does not participate in back-propagation, reducing memory and preventing motion-accumulation error [2511.17059].

TR-Gaussians represent a different planar extension. Here the Gaussians themselves are standard 3DGS primitives with covariance \(\mathbf\Sigma_i=\mathbf R_i\mathbf S_i\mathbf S_i^\top\mathbf R_i^\top\), but the planar structure is introduced through learnable reflection planes \(\mathbf P=(\mathbf u_p,\mathbf n_p,w,h,R_0)\) that mirror Gaussians to synthesize planar reflection and transmission [2511.13009]. The final image is
\[
\mathbf C=(1-\mathbf R)\cdot \mathbf C_t+\mathbf R\cdot \mathbf C_r,
\]
with \(\mathbf R\) given by a Fresnel-based view-dependent term multiplied by a rendered reflection mask [2511.13009]. The paper is explicit that the planar aspect is not that the Gaussians themselves become 2D planar splats; instead, planes act as global operators over a 3D Gaussian set. On RTR, TR-Gaussians reports \(31.21\) PSNR, \(0.951\) SSIM, and \(0.146\) LPIPS, versus 3DGS at \(30.16/0.945/0.158\), while retaining real-time rendering at \(225\) FPS and training in \(33\) minutes [2511.13009].

## 5. Literal 2D planar Gaussians in mapping and vectorization

GSMap uses the term in the most literal geometric sense: a planar Gaussian is a 2D anisotropic kernel on the BEV plane. One Gaussian is
\[
G=(\mu,\sigma,\theta),\qquad \mu=(\mu_x,\mu_y)\in\mathbb R^2,\quad \sigma=(\sigma_x,\sigma_y)\in\mathbb R^2,\quad \theta\in\mathbb R,
\]
with density
\[
G_i(p)=\exp\!\Big(-\tfrac12(p-\mu_i)^\top\mathbf\Sigma_i^{-1}(p-\mu_i)\Big),
\]
and covariance parameterized as
\[
\mathbf\Sigma_i=\mathbf R_i\mathbf S_i\mathbf S_i^\top\mathbf R_i^\top.
\]
Each HD map element is an ordered sequence \(E^j=\{G_1^j,\dots,G_N^j\}\), and the ordered centers \(\{\mu_i^j\}_{i=1}^N\) are connected sequentially to form continuous vectorized map elements [2605.09619].

The central technical point is that the same primitive supports both raster and vector supervision. Rasterization is defined by
\[
\mathcal R_j(p)=1-\prod_{i=1}^N(1-G_i^j(p)),
\]
while vectorization simply reads out the ordered centers. Training combines classification, point regression, and raster loss:
\[
\mathcal L_{\mathrm{ins}}=\mathcal L_{\mathrm{cls}}+\lambda_v\mathcal L_{\mathrm{vector}}+\lambda_r\mathcal L_{\mathrm{raster}}.
\]
The best ablation settings reported are \(\lambda_r=10\), \(N=20\), and rasterization resolution \(200\times 100\) [2605.09619]. Integrated into MapTR, GSMap improves nuScenes val performance at 24 epochs from \(\mathrm{AP}_{\mathrm{Chamfer}}=50.3\), \(\mathrm{AP}_{\mathrm{raster}}=24.3\) for MapTR to \(52.3\) and \(33.9\) for MapTR + GSMap, while retaining near-identical efficiency: \(17.4\) FPS and \(35.95\)M parameters versus \(17.9\) FPS and \(35.92\)M for MapTR [2605.09619].

This literal 2D usage is conceptually important because it separates “planar Gaussian” from “flattened 3D Gaussian.” In GSMap, the Gaussian is not a 3D splat that has been compressed; it is a native 2D primitive whose center doubles as a vector vertex and whose support doubles as a raster occupancy kernel.

## 6. Mathematical meanings: densities, fields, convexity, and planar asymptotics

In mathematics, planar Gaussian terminology refers to the ambient plane or to planar asymptotics, not to surface-oriented splats. The double-Gaussian isoperimetric problem studies the density
\[
f(x,y)=\frac{1}{4\pi a^2}e^{-y^2/(2a^2)}\left(e^{-(x-1)^2/(2a^2)}+e^{-(x+1)^2/(2a^2)}\right)
\]
on \(\mathbb R^2\). The main planar theorem is conditional: among linear boundaries, only horizontal and vertical lines are stationary, and vertical lines enclose given area with less perimeter than horizontal lines; therefore, if planar isoperimetric regions are half-spaces, they must be bounded by vertical lines [1604.00592]. 

A different convex-geometric usage appears in the planar Gaussian-Minkowski problem, where one seeks a convex body whose Gaussian surface area measure matches prescribed data on \(S^1\). For density \(f\,dv\), the support function \(h\) satisfies
\[
\frac{1}{2\pi}e^{-(h'^2+h^2)/2}(h''+h)=f.
\]
For constant data, the paper proves that if the Gaussian surface area measure is proportional to spherical Lebesgue measure, the body must be a centered disk; it then uses this uniqueness result to prove existence of smooth small solutions by a degree-theoretic argument [2303.17389].

Planar Gaussian fields form yet another branch. For smooth stationary planar Gaussian fields, four discretisation schemes recover different levels of topological information about level sets. The mesh conditions are sharply stratified: \(\varepsilon=o(s^{-2})\) for full topology in each face, \(\varepsilon=o(s^{-1-\delta})\) for full topology in \(sD\), \(\varepsilon=o(s^{-1/2})\) for the random plane wave up to visible ambiguities, and \(\varepsilon=o(1)\) for approximating the mean number of excursion domains and hence the Nazarov–Sodin constant [1702.02134]. For stationary planar Gaussian fields with trends, exact formulas are available for the expected number of local maxima and the height distribution of local maxima:
\[
\mathbb E[M(X,T)] = \frac{1}{2\pi}\int_T \exp\!\left(-\frac{m_1(t)^2+m_2(t)^2}{2}\right)J_{1,t}\,dt,
\]
with the Hessian contribution \(J_{1,t}\) reduced to generalized chi-squared densities because GOE-based isotropic machinery no longer applies directly in the non-isotropic, non-centered case [2307.01974].

Finally, “planar” can refer to large-\(N\) asymptotics rather than planar geometry. For Gaussian distributions on Riemannian symmetric spaces, the partition function admits a genus expansion
\[
F(\sigma^2)=\sum_{g=0}^\infty f_g(t)N^{2-2g},
\]
so the large-\(N\) limit is governed by the planar sector \(g=0\). In that sense the paper develops a theory of Gaussian models whose normalization constants are controlled by planar random-matrix and ribbon-graph asymptotics, rather than by Euclidean planar support [2106.08953].

Taken together, these lines of work show that “planar Gaussian” is a technically dense but context-dependent phrase. In reconstruction and rendering it usually means a surface-biased Gaussian primitive, obtained either by collapsing one covariance axis or by constraining Gaussian coordinates to a plane. In mapping it can mean a literal 2D Gaussian on a plane. In mathematical analysis it often means a Gaussian density or field defined on the plane, or a Gaussian partition function whose asymptotics are organized by the planar large-\(N\) limit.

Source: https://www.emergentmind.com/topics/planar-gaussians