mlsanityData to 14 Oct 2026

MuM: Multi-View Masked Image Modeling for 3D Vision

arXiv 2511.17309 ↗First posted 21 Nov 2025cs.CV cs.AI cs.LGcited by 5

Not on the map. Why?

Abstract (arXiv)

Self-supervised learning on images seeks to extract meaningful visual representations from unlabeled data. When scaled to large datasets, this paradigm has achieved state-of-the-art performance and the resulting trained models such as DINOv3 have seen widespread adoption. However, most prior efforts are optimized for semantic understanding rather than geometric reasoning. One important exception is Cross-View Completion, CroCo, which is a form of masked autoencoding (MAE) tailored for 3D understanding. In this work, we continue on the path proposed by CroCo and focus on learning features tailored for 3D vision. In a nutshell, we extend MAE to arbitrarily many views of the same scene. By uniformly masking all views and employing a lightweight decoder with inter-frame attention, our approach is inherently simpler and more scalable than CroCo. We evaluate the resulting model, MuM, extensively on downstream tasks including feedforward reconstruction, dense image matching and relative pose estimation, finding that it outperforms the state-of-the-art visual encoders DINOv3 and CroCo v2.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 57 of its 81 references that are in the map. First posted on arXiv 21 Nov 2025.

Builds on

57 of 81 references.

References · 57

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 30,186
  3. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,498
  4. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  5. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
  6. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,968
  7. 2021RoFormer: Enhanced Transformer with Rotary Position Embeddingcited by 4,703
  8. 2017ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenescited by 3,117
  9. 2021BEiT: BERT Pre-Training of Image Transformerscited by 2,375
  10. 2021Vision Transformers for Dense Predictioncited by 1,803
  11. 2023DUSt3R: Geometric 3D Vision Made Easycited by 1,567
  12. 2019SuperGlue: Learning Feature Matching with Graph Neural Networkscited by 1,443
38 more
  1. 2022VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Trainingcited by 1,322
  2. 2021SimMIM: A Simple Framework for Masked Image Modelingcited by 1,153
  3. 2016Semantic Understanding of Scenes through the ADE20K Datasetcited by 1,069
  4. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 1,039
  5. 2021LoFTR: Detector-Free Local Feature Matching with Transformerscited by 984
  6. 2021iBOT: Image BERT Pre-Training with Online Tokenizercited by 903
  7. 2025V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planningcited by 809
  8. 2023ScanNet++: A High-Fidelity Dataset of 3D Indoor Scenescited by 764
  9. 2018MegaDepth: Learning Single-View Depth Prediction from Internet Photoscited by 737
  10. 2022EVA: Exploring the Limits of Masked Visual Representation Learning at Scalecited by 691
  11. 2021Common Objects in 3D: Large-Scale Learning and Evaluation of Real-life 3D Category Reconstructioncited by 652
  12. 2020Hypersim: A Photorealistic Synthetic Dataset for Holistic Indoor Scene Understandingcited by 604
  13. 2023LightGlue: Local Feature Matching at Light Speedcited by 590
  14. 2023DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Visioncited by 537
  15. 2023VideoMAE V2: Scaling Video Masked Autoencoders with Dual Maskingcited by 528
  16. 2019BlendedMVS: A Large-scale Dataset for Generalized Multi-view Stereo Networkscited by 490
  17. 2025Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Passcited by 367
  18. 2017Sparsity Invariant CNNscited by 326
  19. 2025Perception Encoder: The best visual embeddings are not at the output of the networkcited by 326
  20. 2025π3: Permutation-Equivariant Visual Geometry Learningcited by 319
  21. 2025MapAnything: Universal Feed-Forward Metric 3D Reconstructioncited by 314
  22. 2023RoMa: Robust Dense Feature Matchingcited by 302
  23. 2022MultiMAE: Multi-modal Multi-task Masked Autoencoderscited by 291
  24. 2018DeepMVS: Learning Multi-view Stereopsiscited by 277
  25. 2023PointOdyssey: A Large-Scale Synthetic Dataset for Long-Term Point Trackingcited by 276
  26. 2022DKM: Dense Kernelized Feature Matching for Geometry Estimationcited by 177
  27. 2024Probing the 3D Awareness of Visual Foundation Modelscited by 177
  28. 2022CroCo v2: Improved Cross-view Completion Pre-training for Stereo Matching and Optical Flowcited by 166
  29. 2022CroCo: Self-Supervised Pre-training for 3D Vision Tasks by Cross-View Completioncited by 160
  30. 2024MV-DUSt3R+: Single-Stage Scene Reconstruction from Sparse Views In 2 Secondscited by 141
  31. 2025MUSt3R: Multi-view Network for Stereo 3D Reconstructioncited by 141
  32. 2024Scalable Pre-training of Large Autoregressive Image Modelscited by 115
  33. 2024RGBD Objects in the Wild: Scaling Real-World 3D Object Learning from RGB-D Videoscited by 115
  34. 2024Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localizationcited by 82
  35. 2023Multi-View Masked World Models for Visual Robotic Manipulationcited by 80
  36. 2024MegaScenes: Scene-Level View Synthesis at Scalecited by 58
  37. 2023NAVI: Category-Agnostic Image Collections with High-Quality 3D Shape and Pose Annotationscited by 53
  38. 2025UFM: A Simple Path towards Unified Dense Correspondence with Flowcited by 36

50 most cited of 57 All 57 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026Revisiting Cross-View Completion: Self-Supervised Pre-Training via Reconstruction Error Comparisoncited by 1
  2. 2026RoMa-Omega: What Feed-Forward 3D Models Know About Image Matchingcited by 0
  3. 2026SBMVTrack: Spike-Budgeted Multi-View Learning for Power-Efficient UAV Trackingcited by 0
  4. 2026Emergent Multi-View Geometry Through Self-Distillationcited by 0
  5. 2026Less Decoder is More Encoder: Geometric Representation Learning from Novel View Synthesiscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2022MM-3DScene: 3D Scene Understanding by Customizing Masked Modeling with Informative-Preserved Reconstruction and Self-Distilled Consistency
  2. 2025CDG-MAE: Cross-view Masked Modeling using Diffusion Generated Views
  3. 2026Less Context, Better Geometry: Masked Geometric Encoder for Robust 3D Foundation Models
  4. 2022A simple, efficient and scalable contrastive masked autoencoder for learning visual representations
  5. 2025Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion
  6. 2026Cross-Attentive Multiview Fusion of Vision-Language Embeddings
  7. 2022Autoencoders as Cross-Modal Teachers: Can Pretrained 2D Image Transformers Help 3D Representation Learning?
  8. 2025Ross3D: Reconstructive Visual Instruction Tuning with 3D-Awareness