Reconstruction News
The latest Reconstruction coverage curated for game developers.
arXiv cs.GR details M-plicits, a neural implicit surface method that aims to make reconstruction cleaner, faster, and more noise-tolerant. The system uses nested multiscale residual MLPs plus a sphere-tracing renderer …
arXiv cs.GR details SplashSplat, a new method for reconstructing splashing liquids from real multi-view video. The team pairs a 20-scene benchmark with seven synchronized 4K cameras at 60 fps and …
UMA is a new human-avatar pipeline aimed at producing ultra-detailed digital people through multi-level surface alignment. Published in ACM Transactions on Graphics in February 2027, it targets higher-fidelity body representation …
A new reconstruction method uses a Lipschitz-extension pass to stabilize Moving Least Squares on sparse, irregular samples. The result is better accuracy and fewer failures when data points are unevenly …
ExMesh++ turns multi-view photos into editable UV-PBR mesh assets, aiming to skip the usual mesh extraction and texture-baking detour. The staged pipeline adapts topology with vertex split/merge operations, then locks …
A correction has been issued for the ACM TOG paper on tensor decomposition-based 4D background-oriented schlieren tomography, the kind of work that can influence high-end simulation and visualization pipelines. The …
3D Gaussian splatting is getting a geometry upgrade. A new approach folds predicted normals and depth into the reconstruction pipeline, with multi-view priors proving more reliable than single-view estimates. The …
A new CT reconstruction method built on 3D Gaussian splats is getting a reality check: in sparse-view scans, pose errors are causing stronger streaking and striping than view sparsity itself. …
RORA turns a single static object video into a simulation-ready asset with both photoreal rendering and physical articulation. The pipeline pairs 3D Gaussian Splatting with mesh geometry, then uses automatic …
StructureGS adds structure-aware constraints to 3D Gaussian Splatting for articulated object reconstruction, aiming to keep movable parts compact and physically plausible. The method uses oriented bounding boxes to improve part …
A new point-cloud distance method turns each sample into a locally fitted torus, giving fast signed-distance queries without global optimization or voxelization. For graphics teams, that means point clouds can …
A new 3D agent called CLARE tackles a common production problem: users give vague prompts, but toolchains need exact parameters. It pauses to ask clarifying questions before running expensive 3D …
TopoGS adds topology-aware planar constraints to 3D Gaussian Splatting, aiming to turn fragmented reconstructions into connected scene geometry. The method anchors Gaussians to 2D structural relationships recovered from multi-view segmentations, …
A new splatting pipeline promises up to 5x faster rendering than 3D Gaussian Splatting while keeping state-of-the-art visual quality. It bakes high-frequency detail into a compact texture atlas, then uses …
arXiv cs.GR details SSA-3DGS, an unsupervised method that strips screen-space junk from 3D Gaussian Splatting inputs before it gets baked into the scene. For teams using captured imagery, that means …
A new 3D Gaussian Splatting pipeline skips traditional SfM and gets to usable scenes in about three minutes. It uses 3D foundation models to seed camera poses and point clouds, …
A SIGGRAPH Asia paper shows a way to make specular reflections in Gaussian-splat scenes editable instead of baking in “fake” reflected geometry. The practical win for graphics teams is consistent …
A new differentiable rendering method called Meshtryoshka uses ordinary triangle rasterizers to reconstruct real-world scenes as nested mesh shells. It targets a long-standing gap in graphics research: mesh-based pipelines that …
A new visual analytics tool is tackling one of 3D Gaussian Splatting’s biggest pain points: understanding why reconstruction artifacts appear. Vis4GS links blur and needle-like defects to primitive-level properties, training …
LEGS swaps the usual first-order edge weighting in 3D Gaussian Splatting for Laplacian-based structural guidance, then turns that signal into nonlinear pixel weights. In the paper’s tests, that pushed PSNR …