mlsanityData to 13 Oct 2026

Difix3D+: Improving 3D Reconstructions with Single-Step Diffusion Models

arXiv 2503.01774 ↗First posted 3 Mar 2025cs.CVcited by 160

On the map: Diffusion models → 3D, point cloud and scene generation

Abstract (arXiv)

Neural Radiance Fields and 3D Gaussian Splatting have revolutionized 3D reconstruction and novel-view synthesis task. However, achieving photorealistic rendering from extreme novel viewpoints remains challenging, as artifacts persist across representations. In this work, we introduce Difix3D+, a novel pipeline designed to enhance 3D reconstruction and novel-view synthesis through single-step diffusion models. At the core of our approach is Difix, a single-step image diffusion model trained to enhance and remove artifacts in rendered novel views caused by underconstrained regions of the 3D representation. Difix serves two critical roles in our pipeline. First, it is used during the reconstruction phase to clean up pseudo-training views that are rendered from the reconstruction and then distilled back into 3D. This greatly enhances underconstrained regions and improves the overall 3D representation quality. More importantly, Difix also acts as a neural enhancer during inference, effectively removing residual artifacts arising from imperfect 3D supervision and the limited capacity of current reconstruction models. Difix3D+ is a general solution, a single model compatible with both NeRF and 3DGS representations, and it achieves an average 2× improvement in FID score over baselines while maintaining 3D consistency.

Key facts. Cited by 160 papers in the map (data through 13 Oct 2026). Placed in 3D, point cloud and scene generation, one of 14 directions in Diffusion models. Builds on 78 of its 88 references that are in the map. First posted on arXiv 3 Mar 2025.

Builds on

78 of 88 references.

Closest ancestors · 16

Same direction, newest first

  1. 2024SCube: Instant Large-Scale Scene Reconstruction using VoxSplats26 Oct
  2. 20243DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with View-consistent 2D Diffusion Priors21 Oct
  3. 2024ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis3 Sep
  4. 2024CAT3D: Create Anything in 3D with Multi-View Diffusion Models16 May
  5. 2023ReconFusion: 3D Reconstruction with Diffusion Priors5 Dec
  6. 2023One-2-3-45++: Fast Single Image to 3D Objects with Consistent Multi-View Generation and 3D Diffusion14 Nov
  7. 2023Zero123++: a Single Image to Consistent Multi-view Diffusion Base Model23 Oct
  8. 2023Wonder3D: Single Image to 3D using Cross-Domain Diffusion23 Oct
  9. 2023Gradpaint: Gradient-Guided Inpainting with Diffusion Models18 Sep
  10. 2023SyncDreamer: Generating Multiview-consistent Images from a Single-view Image7 Sep
  11. 2023MVDream: Multi-view Diffusion for 3D Generation31 Aug
  12. 2023Magic123: One Image to High-Quality 3D Object Generation Using Both 2D and 3D Diffusion Priors30 Jun
4 more
  1. 2023Long-Term Photometric Consistent Novel View Synthesis with Diffusion Models21 Apr
  2. 2023Towards Coherent Image Inpainting Using Denoising Diffusion Implicit Models6 Apr
  3. 2023Zero-1-to-3: Zero-shot One Image to 3D Object20 Mar
  4. 2022DreamFusion: Text-to-3D using 2D Diffusion29 Sep

Other references · 62

  1. 2020Denoising Diffusion Probabilistic Modelscited by 20,952
  2. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,321
  3. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,028
  4. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,023
  5. 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,170
  6. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,077
  7. 2016Perceptual Losses for Real-Time Style Transfer and Super-Resolutioncited by 2,687
  8. 2021Mip-NeRF 360: Unbounded Anti-Aliased Neural Radiance Fieldscited by 2,021
  9. 2022Imagen Video: High Definition Video Generation with Diffusion Modelscited by 1,768
  10. 2022Make-A-Video: Text-to-Video Generation without Text-Video Datacited by 1,732
  11. 2020pixelNeRF: Neural Radiance Fields from One or Few Imagescited by 1,581
  12. 2023Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Modelscited by 1,467
38 more
  1. 2023AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuningcited by 1,422
  2. 2020NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collectionscited by 1,231
  3. 2023One-step Diffusion with Distribution Matching Distillationcited by 991
  4. 2022Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generationcited by 959
  5. 2020NeRF++: Analyzing and Improving Neural Radiance Fieldscited by 858
  6. 2025Cosmos World Foundation Model Platform for Physical AIcited by 848
  7. 2023Latent Consistency Models: Synthesizing High-Resolution Images with Few-Step Inferencecited by 837
  8. 2023Adversarial Diffusion Distillationcited by 751
  9. 2021MVSNeRF: Fast Generalizable Radiance Field Reconstruction from Multi-View Stereocited by 738
  10. 2021Depth-supervised NeRF: Fewer Views and Faster Training for Freecited by 733
  11. 2019Local Light Field Fusion: Practical View Synthesis with Prescriptive Sampling Guidelinescited by 672
  12. 2024Improved Distribution Matching Distillation for Fast Image Synthesiscited by 664
  13. 2021BARF: Bundle-Adjusting Neural Radiance Fieldscited by 641
  14. 2021RegNeRF: Regularizing Neural Radiance Fields for View Synthesis from Sparse Inputscited by 614
  15. 2023Nerfstudio: A Modular Framework for Neural Radiance Field Developmentcited by 597
  16. 2023DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Visioncited by 521
  17. 2023Instruct-NeRF2NeRF: Editing 3D Scenes with Instructionscited by 483
  18. 2022MonoSDF: Exploring Monocular Geometric Cues for Neural Implicit Surface Reconstructioncited by 468
  19. 2022MagicVideo: Efficient Video Generation With Latent Diffusion Modelscited by 465
  20. 2023LAVIE: High-Quality Video Generation with Cascaded Latent Diffusion Modelscited by 349
  21. 2023FreeNeRF: Improving Few-shot Neural Rendering with Free Frequency Regularizationcited by 310
  22. 2021Dense Depth Priors for Neural Radiance Fields from Sparse Input Viewscited by 301
  23. 2023Show-1: Marrying Pixel and Latent Diffusion Models for Text-to-Video Generationcited by 301
  24. 2023Preserve Your Own Correlation: A Noise Prior for Video Diffusion Modelscited by 285
  25. 2023FSGS: Real-Time Few-shot View Synthesis using Gaussian Splattingcited by 279
  26. 2024SDXL-Lightning: Progressive Adversarial Diffusion Distillationcited by 261
  27. 2023Emu Video: Factorizing Text-to-Video Generation by Explicit Image Conditioningcited by 254
  28. 2023SparseNeRF: Distilling Depth Ranking for Few-shot Novel View Synthesiscited by 247
  29. 2022Imagen Editor and EditBench: Advancing and Evaluating Text-Guided Image Inpaintingcited by 231
  30. 2022SparseFusion: Distilling View-conditioned Diffusion for 3D Reconstructioncited by 216
  31. 2023DMV3D: Denoising Multi-View Diffusion using 3D Large Reconstruction Modelcited by 213
  32. 2022SPARF: Neural Radiance Fields from Sparse and Noisy Posescited by 195
  33. 2023NerfDiff: Single-image View Synthesis with NeRF-guided Distillation from 3D-aware Diffusioncited by 180
  34. 2023Swap Attention in Spatiotemporal Diffusions for Text-to-Video Generationcited by 154
  35. 2022FILM: Frame Interpolation for Large Motioncited by 151
  36. 2023SUDS: Scalable Urban Dynamic Scenescited by 151
  37. 2024One-Step Image Translation with Text-to-Image Modelscited by 113
  38. 2023Progressively Optimized Local Radiance Fields for Robust View Synthesiscited by 103

50 most cited of 62 All 62 (JSON)

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2025: 52; 2026: 108.

160 papers.

Sort:

By area:

  1. 2025WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modelingcited by 158
  2. 2025Pseudo-Simulation for Autonomous Drivingcited by 142
  3. 2025Cosmos-Drive-Dreams: Scalable Synthetic Driving Data Generation with World Foundation Modelscited by 81
  4. 2025GigaWorld-0: World Models as Data Engine to Empower Embodied AIcited by 62
  5. 2026SANA-WM: Efficient Minute-Scale World Modeling with Hybrid Linear Diffusion Transformercited by 52
  6. 2026NeoVerse: Enhancing 4D World Model with in-the-wild Monocular Videoscited by 41
  7. 2025GSFixer: Improving 3D Gaussian Splatting with Reference-Guided Video Diffusion Priorscited by 37
  8. 2025Advances in Feed-Forward 3D Reconstruction and View Synthesis: A Surveycited by 25
  9. 2025RealX3D: A Physically-Degraded 3D Benchmark for Multi-view Visual Restoration and Reconstructioncited by 19
  10. 2025StochasticSplats: Stochastic Rasterization for Sorting-Free 3D Gaussian Splattingcited by 18
  11. 2025R3D2: Realistic 3D Asset Insertion via Diffusion for Autonomous Driving Simulationcited by 17
  12. 2025GSFix3D: Diffusion-Guided Repair of Novel Views in Gaussian Splattingcited by 17
38 more
  1. 2025ORV: 4D Occupancy-centric Robot Video Generationcited by 16
  2. 2025Taming Video Models for 3D and 4D Generation via Zero-Shot Camera Controlcited by 15
  3. 2025Quantifying and Alleviating Co-Adaptation in Sparse-View 3D Gaussian Splattingcited by 13
  4. 2025DualCamCtrl: Dual-Branch Diffusion Model for Geometry-Aware Camera-Controlled Video Generationcited by 11
  5. 2026NTIRE 2026 3D Restoration and Reconstruction in Real-world Adverse Conditions: RealX3D Challenge Resultscited by 11
  6. 2025G4Splat: Geometry-Guided Gaussian Splatting with Generative Priorcited by 10
  7. 2026One2Scene: Geometric Consistent Explorable 3D Scene Generation from a Single Imagecited by 10
  8. 2026DiffusionHarmonizer: Bridging Neural Reconstruction and Photorealistic Simulation with Online Diffusion Enhancercited by 10
  9. 2026World Reconstruction From Inconsistent Viewscited by 8
  10. 20254DSloMo: 4D Reconstruction for High Speed Scene with Asynchronous Capturecited by 6
  11. 2026ArtiFixer: Enhancing and Extending 3D Reconstruction with Auto-Regressive Diffusion Modelscited by 6
  12. 2026GaussFusion: Improving 3D Reconstruction in the Wild with A Geometry-Informed Video Generatorcited by 6
  13. 2025SeeU: Seeing the Unseen World via 4D Dynamics-aware Generationcited by 5
  14. 2025From Orbit to Ground: Generative City Photogrammetry from Extreme Off-Nadir Satellite Imagescited by 5
  15. 2026GSMem: 3D Gaussian Splatting as Persistent Spatial Memory for Zero-Shot Embodied Exploration and Reasoningcited by 5
  16. 2026ArtifactWorld: Scaling 3D Gaussian Splatting Artifact Restoration via Video Generation Modelscited by 5
  17. 2026Xiaomi Auto World Model: A Joint World Model Integrating Reconstruction and Generation for Autonomous Drivingcited by 5
  18. 2025ViDAR: Video Diffusion-Aware 4D Reconstruction From Monocular Inputscited by 4
  19. 2025ExtraGS: Geometric-Aware Trajectory Extrapolation with Uncertainty-Guided Generative Priorscited by 4
  20. 2025GaMO: Geometry-aware Multi-view Diffusion Outpainting for Sparse-View 3D Reconstructioncited by 4
  21. 2026GIFSplat: Generative Prior-Guided Iterative Feed-Forward 3D Gaussian Splatting from Sparse Viewscited by 4
  22. 2026SyncFix: Fixing 3D Reconstructions via Multi-View Synchronizationcited by 4
  23. 2026AnyRecon: Arbitrary-View 3D Reconstruction with Video Diffusion Modelcited by 4
  24. 2025ProSplat: Improved Feed-Forward 3D Gaussian Splatting for Wide-Baseline Sparse Viewscited by 3
  25. 2025MADrive: Memory-Augmented Driving Scene Modelingcited by 3
  26. 2025Inference-time Scaling for Diffusion-based Audio Super-resolutioncited by 3
  27. 2025FixingGS: Enhancing 3D Gaussian Splatting via Training-Free Score Distillationcited by 3
  28. 2025SimULi: Real-Time LiDAR and Camera Simulation with Unscented Transformscited by 3
  29. 2025Wanderland: Geometrically Grounded Simulation for Open-World Embodied AIcited by 3
  30. 2025Breaking the Vicious Cycle: Coherent 3D Gaussian Splatting from Sparse and Motion-Blurred Viewscited by 3
  31. 2025LangDriveCTRL: Natural Language Controllable Driving Scene Editing with Multi-modal Agentscited by 3
  32. 2025Mirage: One-Step Video Diffusion for Photorealistic and Coherent Asset Editing in Driving Scenescited by 3
  33. 2026HorizonForge: Driving Scene Editing with Any Trajectories and Any Vehiclescited by 3
  34. 2026GeoNVS: Geometry Grounded Video Diffusion for Novel View Synthesiscited by 3
  35. 2026From Articulated Kinematics to Routed Visual Control for Action-Conditioned Surgical Video Generationcited by 3
  36. 2026TryOnCrafter: Unleashing Camera Trajectories for Realistic Video Virtual Try-on via a Renderable 4D Try-on Proxycited by 3
  37. 2025BulletGen: Improving 4D Reconstruction with Bullet-Time Generationcited by 2
  38. 2025WaveletGaussian: Wavelet-domain Diffusion for Sparse-view 3D Gaussian Object Reconstructioncited by 2

50 most cited of 160 All 160 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2025Diffusion-Guided Gaussian Splatting for Large-Scale Unconstrained 3D Reconstruction and Novel View Synthesis
  2. 2022DiffRF: Rendering-Guided 3D Radiance Field Diffusion
  3. 2023Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views
  4. 2023Single-Stage Diffusion NeRF: A Unified Approach to 3D Generation and Reconstruction
  5. 2025DT-NeRF: A Diffusion and Transformer-Based Optimization Approach for Neural Radiance Fields in 3D Reconstruction
  6. 2025Diff4Splat: Controllable 4D Scene Generation with Latent Dynamic Reconstruction Models
  7. 2023Deceptive-NeRF/3DGS: Diffusion-Generated Pseudo-Observations for High-Quality Sparse-View Reconstruction
  8. 2023Efficient-3DiM: Learning a Generalizable Single-image Novel-view Synthesizer in One Day