mlsanityData to 14 Oct 2026

Off The Grid: Detection of Primitives for Feed-Forward 3D Gaussian Splatting

arXiv 2512.15508 ↗First posted 17 Dec 2025cs.CVcited by 11

Not on the map. Why?

Abstract (arXiv)

Feed-forward 3D Gaussian Splatting (3DGS) models enable real-time scene generation but are hindered by suboptimal pixel-aligned primitive placement, which relies on a dense, rigid grid that limits both quality and efficiency. We introduce a new feed-forward architecture that detects 3D Gaussian primitives at a sub-pixel level, replacing the pixel grid with an adaptive, ``Off-The-Grid" distribution. Inspired by keypoint detection, our decoder learns to locally distribute primitives across image patches. We also provide an Adaptive Density mechanism by assigning varying number of primitives per patch based on Shannon entropy. We combine the proposed decoder with a pre-trained 3D reconstruction backbone and train them end-to-end using photometric supervision without any 3D annotation. The resulting pose-free model generates photorealistic 3DGS scenes in seconds, achieving state-of-the-art novel view synthesis for feed-forward models. It outperforms competitors while using far fewer primitives, demonstrating a more accurate and efficient allocation that captures fine details and reduces artifacts. Project page: https://arthurmoreau.github.io/OffTheGrid/.

Key facts. Cited by 11 papers in the map (data through 14 Oct 2026). Builds on 62 of its 74 references that are in the map. First posted on arXiv 17 Dec 2025.

Builds on

62 of 74 references.

References · 62

  1. 2015U-Net: Convolutional Networks for Biomedical Image Segmentationcited by 19,713
  2. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  3. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  4. 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,226
  5. 2022Instant Neural Graphics Primitives with a Multiresolution Hash Encodingcited by 3,723
  6. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,117
  7. 2021Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fieldscited by 2,032
  8. 2023DUSt3R: Geometric 3D Vision Made Easycited by 1,567
  9. 2025VGGT: Visual Geometry Grounded Transformercited by 1,539
  10. 2017SuperPoint: Self-Supervised Interest Point Detection and Descriptioncited by 1,320
  11. 2024Grounding Image Matching in 3D with MASt3Rcited by 983
  12. 2025Depth Anything 3: Recovering the Visual Space from Any Viewscited by 692
38 more
  1. 2023pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstructioncited by 684
  2. 2021Common Objects in 3D: Large-Scale Learning and Evaluation of Real-life 3D Category Reconstructioncited by 652
  3. 2021KITTI-360: A Novel Dataset and Benchmarks for Urban Scene Understanding in 2D and 3Dcited by 575
  4. 2024MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Imagescited by 566
  5. 2023DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Visioncited by 537
  6. 2019BlendedMVS: A Large-scale Dataset for Generalized Multi-view Stereo Networkscited by 490
  7. 2021ARKitScenes: A Diverse Real-World Dataset For 3D Indoor Scene Understanding Using Mobile RGB-D Datacited by 454
  8. 2023Gaussian Splatting SLAMcited by 415
  9. 2023Splatter Image: Ultra-Fast Single-View 3D Reconstructioncited by 364
  10. 2024GS-LRM: Large Reconstruction Model for 3D Gaussian Splattingcited by 349
  11. 2025π3: Permutation-Equivariant Visual Geometry Learningcited by 319
  12. 2025MapAnything: Universal Feed-Forward Metric 3D Reconstructioncited by 314
  13. 2024DepthSplat: Connecting Gaussian Splatting and Depthcited by 286
  14. 2024No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse Unposed Imagescited by 248
  15. 2025AnySplat: Feed-forward 3D Gaussian Splatting from Unconstrained Viewscited by 243
  16. 2024Taming 3DGS: High-Quality Radiance Fields with Limited Resourcescited by 204
  17. 2023COLMAP-Free 3D Gaussian Splattingcited by 193
  18. 2024LVSM: A Large View Synthesis Model with Minimal 3D Inductive Biascited by 189
  19. 2025FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Viewscited by 180
  20. 2024Splatt3R: Zero-shot Gaussian Splatting from Uncalibrated Image Pairscited by 178
  21. 2024Revising Densification in Gaussian Splattingcited by 128
  22. 2021SMD-Nets: Stereo Mixture Density Networkscited by 120
  23. 2024MVSplat360: Feed-Forward 360 Scene Synthesis from Sparse Viewscited by 116
  24. 2024RGBD Objects in the Wild: Scaling Real-World 3D Object Learning from RGB-D Videoscited by 115
  25. 2024RaDe-GS: Rasterizing Depth in Gaussian Splattingcited by 111
  26. 2023Human Gaussian Splatting: Real-time Rendering of Animatable Avatarscited by 105
  27. 2024FreeSplat: Generalizable 3D Gaussian Splatting Towards Free-View Synthesis of Indoor Scenescited by 89
  28. 2025RayZer: A Self-supervised Large View Synthesis Modelcited by 77
  29. 2018Numerical Coordinate Regression with Convolutional Neural Networkscited by 72
  30. 2023HiFi4G: High-Fidelity Human Performance Rendering via Compact Gaussian Splattingcited by 70
  31. 2024InstantSplat: Sparse-view Gaussian Splatting in Secondscited by 70
  32. 2024PF3plat: Pose-Free Feed-Forward 3D Gaussian Splattingcited by 62
  33. 20254DGT: Learning a 4D Gaussian Transformer Using Real-World Monocular Videoscited by 49
  34. 2025VolSplat: Rethinking Feed-Forward 3D Gaussian Splatting with Voxel-Aligned Predictioncited by 48
  35. 2025FastGS: Training 3D Gaussian Splatting in 100 Secondscited by 43
  36. 2024SelfSplat: Pose-Free and 3D Prior-Free Generalizable 3D Gaussian Splattingcited by 42
  37. 2025No Pose at All: Self-Supervised Pose-Free 3D Gaussian Splatting from Sparse Viewscited by 40
  38. 2025DashGaussian: Optimizing 3D Gaussian Splatting in 200 Secondscited by 38

50 most cited of 62 All 62 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 11.

11 papers.

  1. 2026Feed-Forward 3D Scene Modeling: A Problem-Driven Perspectivecited by 6
  2. 2026Face Anything: 4D Face Reconstruction from Any Image Sequencecited by 2
  3. 2026SplatWeaver: Learning to Allocate Gaussian Primitives for Generalizable Novel View Synthesiscited by 2
  4. 2026Learn2Splat: Extending the Horizon of Learned 3DGS Optimizationcited by 1
  5. 2026Wild3R: Feed-Forward 3D Gaussian Splatting from Unconstrained Sparse Photo Collectioncited by 1
  6. 2026ComPose: When to Trust Hands for Object Pose Trackingcited by 0
  7. 2026G2SR: Geometric Methods for Fast and Memory-Efficient Gaussian-based Surface Reconstructioncited by 0
  8. 2026Compact Feed-Forward 3D Gaussians via Saliency-Guided Primitive Mergingcited by 0
  9. 2026PhGS: Post-Hoc Pruning and Refinement of Single-View Feed-Forward 3D Gaussian Reconstructionscited by 0
  10. 2026PocketSplat: Mobile Gaussian Reconstruction via World-Space Latent Allocatiocited by 0
  11. 2026CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstructioncited by 0