mlsanityData to 13 Oct 2026

UFO-4D: Unposed Feedforward 4D Reconstruction from Two Images

arXiv 2602.24290 ↗First posted 27 Feb 2026cs.CVcited by 4

Not on the map. Why?

Abstract (arXiv)

Dense 4D reconstruction from unposed images remains a critical challenge, with current methods relying on slow test-time optimization or fragmented, task-specific feedforward models. We introduce UFO-4D, a unified feedforward framework to reconstruct a dense, explicit 4D representation from just a pair of unposed images. UFO-4D directly estimates dynamic 3D Gaussian Splats, enabling the joint and consistent estimation of 3D geometry, 3D motion, and camera pose in a feedforward manner. Our core insight is that differentiably rendering multiple signals from a single Dynamic 3D Gaussian representation offers major training advantages. This approach enables a self-supervised image synthesis loss while tightly coupling appearance, depth, and motion. Since all modalities share the same geometric primitives, supervising one inherently regularizes and improves the others. This synergy overcomes data scarcity, allowing UFO-4D to outperform prior work by up to 3 times in joint geometry, motion, and camera pose estimation. Our representation also enables high-fidelity 4D interpolation across novel views and time. Please visit our project page for visual results: https://ufo-4d.github.io/

Key facts. Cited by 4 papers in the map (data through 13 Oct 2026). Builds on 52 of its 65 references that are in the map. First posted on arXiv 27 Feb 2026.

Builds on

52 of 65 references.

References · 52

  1. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,043
  2. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,082
  3. 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,170
  4. 2021Vision Transformers for Dense Predictioncited by 1,795
  5. 2023DUSt3R: Geometric 3D Vision Made Easycited by 1,539
  6. 2025VGGT: Visual Geometry Grounded Transformercited by 1,497
  7. 2024Grounding Image Matching in 3D with MASt3Rcited by 965
  8. 2016Unsupervised Monocular Depth Estimation with Left-Right Consistencycited by 819
  9. 2023Dynamic 3D Gaussians: Tracking by Persistent Dynamic View Synthesiscited by 735
  10. 2021DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Camerascited by 676
  11. 2025Continuous 3D Perception Model with Persistent Statecited by 534
  12. 2024MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motioncited by 437
38 more
  1. 2023Gaussian Splatting SLAMcited by 410
  2. 2016Virtual Worlds as Proxy for Multi-Object Tracking Analysiscited by 341
  3. 2023PointOdyssey: A Large-Scale Synthetic Dataset for Long-Term Point Trackingcited by 274
  4. 20243D Reconstruction with Spatial Memorycited by 271
  5. 2024Shape of Motion: 4D Reconstruction from a Single Videocited by 248
  6. 2024No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse Unposed Imagescited by 243
  7. 2023Visual Geometry Grounded Deep Structure From Motioncited by 226
  8. 2024Global Structure-from-Motion Revisitedcited by 215
  9. 2021Pixel-Perfect Structure-from-Motion with Featuremetric Refinementcited by 200
  10. 2023Tracking Everything Everywhere All at Oncecited by 191
  11. 2019ReFusion: 3D Reconstruction in Dynamic Environments for RGB-D Cameras Exploiting Residualscited by 180
  12. 2024MoSca: Dynamic Gaussian Fusion from Casual Videos via 4D Motion Scaffoldscited by 178
  13. 2025FLARE: Feed-forward Geometry, Appearance and Camera Estimation from Uncalibrated Sparse Viewscited by 176
  14. 2023NICER-SLAM: Neural Implicit Scene Encoding for RGB SLAMcited by 163
  15. 2024SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistencycited by 143
  16. 2024Flow as the Cross-Domain Manipulation Interfacecited by 127
  17. 2021TransMVSNet: Global Context-aware Multi-view Stereo Network with Transformerscited by 117
  18. 2024Diffusion4D: Fast Spatial-temporal Consistent 4D Generation via Video Diffusion Modelscited by 100
  19. 2024Stereo4D: Learning How Things Move in 3D from Internet Stereo Videoscited by 89
  20. 2025Easi3R: Estimating Disentangled Motion from DUSt3R Without Trainingcited by 89
  21. 2025St4RTrack: Simultaneous 4D Reconstruction and Tracking in the Worldcited by 89
  22. 2024Align3R: Aligned Monocular Depth Estimation for Dynamic Videoscited by 79
  23. 2024Deep Patch Visual SLAMcited by 69
  24. 20244Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Modelscited by 64
  25. 2024MoDGS: Dynamic Gaussian Splatting from Casually-captured Monocular Videos with Depth Priorscited by 51
  26. 2019Track to Reconstruct and Reconstruct to Trackcited by 49
  27. 2024Feed-Forward Bullet-Time Reconstruction of Dynamic Scenes from Monocular Videoscited by 48
  28. 2025Light3R-SfM: Towards Feed-forward Structure-from-Motioncited by 47
  29. 2021Multi-View Depth Estimation by Fusing Single-View Depth Probability with Multi-View Geometrycited by 46
  30. 20254DGT: Learning a 4D Gaussian Transformer Using Real-World Monocular Videoscited by 46
  31. 2025Dynamic Point Maps: A Versatile Representation for Dynamic 3D Reconstructioncited by 44
  32. 2025MoVieS: Motion-Aware 4D Dynamic View Synthesis in One Secondcited by 33
  33. 2024Cross-View Completion Models are Zero-shot Correspondence Estimatorscited by 30
  34. 2017Monocular Dense 3D Reconstruction of a Complex Dynamic Scene from Two Perspective Framescited by 27
  35. 2016Temporally coherent 4D reconstruction of complex dynamic scenescited by 21
  36. 2025Zero-Shot Monocular Scene Flow Estimation in the Wildcited by 21
  37. 2025C3G: Learning Compact 3D Representations with 2K Gaussianscited by 18
  38. 2025DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videoscited by 14

50 most cited of 52 All 52 (JSON)

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026No Pose, No Problem in 4D: Feed-Forward Dynamic Gaussians from Unposed Multi-View Videoscited by 2
  2. 2026Learning Global Motion with Compact Gaussians for Feed-Forward 4D Reconstructioncited by 2
  3. 2026Envision4D: Envisioning Visual Futures via Feed-forward 4D Gaussian Splatting for Autonomous Drivingcited by 1
  4. 2026RVLoss: Runoff Vote Loss for Self-Supervised LiDAR Scene Flow Estimationcited by 0