In production systems, 3D geometry is represented by primitives such as points, lines, and triangles, as rendering hardware is optimized to generate images from them very efficiently. Learning such primitives from images by volume rendering is challenging, as they are Dirac masses defined at infinitesimal locations in space. EWA splatting, Poisson reconstruction, and 3D Gaussian Splatting overcome this by convolving Dirac masses with a low-frequency kernel, a recipe that generalizes along two axes: the kernel and the primitive. As points and triangles have already received significant attention, we investigate this recipe for curvilinear primitives, and introduce Spaghetti Splatting. We approximate curves by polylines whose convolution converges to that of the continuous curve, and volume render each segment so that connected segments composite without overcounting at their joints. For fast training and inference, we linearize the camera projection per segment, which also accommodates kernels beyond Gaussians, such as Beta kernels. As a point is a curve of zero length, our representation is a strict superset of 3D Gaussian Splatting. It remains competitive with the state-of-the-art on general scenes, outperforms methods specialized to line-like geometry, natively supports anti-aliasing, and produces long connected strands that are directly amenable to physics simulation.
Figures & tables
Figure 2: Kernel ⊛ dirac – Since Dirac masses and convolution kernels are independent degrees of freedom, we can modify either one to obtain a different differentiable primitive. The first column shows that convolving a point mass with a Gaussian or a Beta kernel yields the primitives of 3DGS [ 13 ] and DBS [ 16 ] , respectively. Replacing the point mass with a line segment, while retaining either kernel, yields the primitives in the middle. Finally, we connect these segments into a polyline, which yields our curvilinear primitive. In the bottom row, we convolve the same Dirac masses with an anisotropic Beta kernel. For visualization, we shade the primitives with the normals of their convolution surfaces [ 4 ] .
Figure 3: Inside our spaghetti – We visualize the Dirac measure of Kheradmand et al. [14] (left), and of our Spaghetti Splatting (right) for the Bonsai scene from Mip-NeRF 360 [ 2 ] . (middle) We show one RGB render, not repeated, as it is visually identical for both representations. While 3DGS models thin strokes with scattered points (e.g. the bicycle spokes), our segments form continuous strokes that follow edges.
Figure 4: Overview – (Left) Our representation is a set of learnable polyline primitives convolved with Gaussian or Beta kernels. (Middle) Without loss of generality, we render each segment individually: we first project it onto the image plane, and then evaluate the integral of the convolution along the ray in screen space using the closed-form integrals of Section 3.3 , which yields a per-segment screen-space rendering. The segments are then composited along each ray to obtain the accumulated color. (Right) We compute the L1 loss between the rendered image and the ground-truth image, and backpropagate its gradient to the parameters of the primitives.
Figure 5: Convergence – A curve rendered as a polyline of 4, 8, 16, and 32 Beta segments rasterized independently converges to the smooth curve without knots at the joints.
Figure 6: Densification – We take inspiration from the densification strategy of 3DGS MCMC [ 14 ] , but adapt it to our primitives. When the opacity of a segment falls below a threshold, MCMC relocation moves it, potentially breaking up the polyline it was in. The relocation target is then sampled in two ways. Long segments are split into two, short segments are cloned.
Figure 7: Comparison on Mip-NeRF 360, Shelly, and Fuzzy – (a, b) Unlike point-based primitives [ 14 , 29 , 8 ] , our method can reconstruct thin elongated geometry effectively. (c) We can also model wide texture-less areas, like the dark shadow on the figure, with wide kernels, unlike Line Primitives [ 34 ] . We mark the three best values in each column with gold, silver, and bronze medals.
Figure 8: Comparison with Gaussian Haircut [ 43 ] – averaged over two datasets: GeomHair [ 15 ] (selected 8 hairstyles) and the curly hair model of Yuksel [42] . Top: strands of the curly hair model in a held-out view, one random color per strand. Our strands closely match the ground-truth strands, whereas those of Gaussian Haircut are poorly structured.
Midpoint split
Segment cloning
Polyline merging
Attraction loss
PSNR ↑
Mean connected component size
✗
✓
✓
✓
33.13
3.26
✓
✗
✓
✓
32.54
4.00
✓
✓
✗
✓
33.59
1.00
✓
✓
✓
✗
33.57
2.77
✓
✓
✓
✓
33.54
3.50
Table 1: Ablations – Averaged over Fuzzy, Shelly, and Mip-NeRF 360; each row removes one component ( ✗ ). The size of a connected component is its number of segments; a segment without a joint has size 1.
While feed-forward 3D Gaussian Splatting (3DGS) enables efficient 3D reconstruction, achieving high-fidelity rendering remains challenging. Existing pixel-aligned approaches suffer from spatial inflexibility and massive structural redundancy, whereas query-based methods lack 3D priors and entangle geometry with appearance, yielding blurry, pose-dependent results. To overcome these deficiencies, we propose QuerySplat, a feed-forward 3DGS framework driven by geometric priors and explicit appearance decoupling. Specifically, we design a dual-branch query-based decoder: the geometry branch leverages a pretrained Vision Geometric Model for spatial understanding, which intrinsically endows QuerySplat with pose-free modeling capabilities, while the appearance branch recovers high-frequency details through a dedicated pathway separated from geometric attribute regression. Extensive experiments demonstrate that QuerySplat mitigates the blurry rendering issues of early query-based models and consistently outperforms pixel-aligned approaches in rendering fidelity. On the challenging DL3DV benchmark, it achieves state-of-the-art novel view synthesis performance, with average PSNR gains of 2.30 dB and 1.04 dB over the best pose-free and pose-required baselines, respectively. Project Page: https://inspatio.github.io/querysplat.
Yinglong Li, Donghui Shen, Xiaoyu Zhang +5
State Key Laboratory of Virtual Reality Technology and Systems, Beihang University · InSpatio Research · State Key Lab of CAD&CG, Zhejiang University
Existing pruning methods for 3D Gaussian splatting (3DGS) suffer from either severe quality degradation or prohibitive computational overhead. In this paper, we propose REFINE, a highly accelerated 3DGS pruning framework centered on a novel rendering-free primitive importance metric. Our approach leverages an analytically approximated, rendering-aware Hessian field to quantify the expected perceptual error induced by the removal of individual primitives. By modeling the joint modulation of visibility, projection geometry and the content adaptive hyperparameter, we entirely bypass costly forward rendering passes and derive an anisotropic perceptual weight field that serves as a high-fidelity proxy for primitive importance. Extensive experiments across multiple benchmark datasets demonstrate that REFINE maintains highly competitive rendering quality while achieving a 3,000× reduction in pruning-related computational complexity, translating to a practical ∼20× speedup in device latency compared to state-of-the-art pruning methods.
Zhang Chen, Shuai Wan, Mengting Yu +2
School of Electronics and Information, Northwestern Polytechnical University · School of Telecommunication Engineering, Xidian University · Department of Computer Science, City University of Hong Kong
We present VkSplat, a high-performance, cross-vendor 3D Gaussian Splatting (3DGS) training pipeline implemented fully in Vulkan compute, addressing performance and compatibility limitation of existing training pipelines. With various optimizations, we achieve 3.3× speed and 33% VRAM reduction over CUDA+PyTorch baseline, maintaining quality, and demonstrating compatibility across GPU vendors. To the best of our knowledge, this is the first fully-Vulkan-based 3DGS training pipeline that achieves state-of-the-art performance. Code: https://github.com/harry7557558/vksplat