mlsanityData to 14 Oct 2026

UniverSat: Resolution- and Modality-Agnostic Transformers for Earth Observation

arXiv 2606.23503 ↗First posted 22 Jun 2026cs.CVcited by 5

Not on the map. Why?

Abstract (arXiv)

Vision Transformers (ViT) dominate computer vision. However, their reliance on rigid patch projectors hinders transfer to Earth Observation (EO), where input modalities, scales, and resolutions vary widely. We introduce UniverSat, a ViT-style backbone built around a Universal Patch Encoder that maps patches from arbitrary spatial, spectral, and temporal resolutions, and from both optical and non-optical sensors, into a shared embedding space with a shared set of weights. This enables training a single model on heterogeneous multimodal corpora via self-supervision, yielding robust, sensor-agnostic spatial features. We validate this approach with strong results across classification and segmentation on standard EO benchmarks from GeoBench, PANGEABench, and SpectralEarth. Our code and models are available at https://github.com/gastruc/UniverSat.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 53 of its 65 references that are in the map. First posted on arXiv 22 Jun 2026.

Builds on

53 of 65 references.

References · 53

  1. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,974
  2. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,513
  3. 2021Masked Autoencoders Are Scalable Vision Learnerscited by 7,403
  4. 2021Emerging Properties in Self-Supervised Vision Transformerscited by 6,968
  5. 2018Unified Perceptual Parsing for Scene Understandingcited by 1,271
  6. 2023Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecturecited by 1,039
  7. 2023Vision Transformers Need Registerscited by 871
  8. 2022SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagerycited by 408
  9. 2019Axial Attention in Multidimensional Transformerscited by 358
  10. 2022Scale-MAE: A Scale-Aware Masked Autoencoder for Multiscale Geospatial Representation Learningcited by 308
  11. 2025Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Freecited by 283
  12. 2024Rotary Position Embedding for Vision Transformercited by 236
38 more
  1. 2021Seasonal Contrast: Unsupervised Pre-Training from Uncurated Remote Sensing Datacited by 215
  2. 2020Geography-Aware Self-Supervised Learningcited by 190
  3. 2023CROMA: Remote Sensing Representations with Contrastive Radar-Optical Masked Autoencoderscited by 169
  4. 2022SatlasPretrain: A Large-Scale Dataset for Remote Sensing Image Understandingcited by 167
  5. 2022FlexiViT: One Model for All Patch Sizescited by 152
  6. 2024Prithvi-EO-2.0: A Versatile Multi-Temporal Foundation Model for Earth Observation Applicationscited by 144
  7. 2021Panoptic Segmentation of Satellite Image Time Series with Convolutional Temporal Attention Networkscited by 143
  8. 2022Self-supervised Learning in Remote Sensing: A Reviewcited by 137
  9. 2025AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label datacited by 137
  10. 2023GEO-Bench: Toward Foundation Models for Earth Monitoringcited by 124
  11. 2023Towards Geospatial Foundation Models via Continual Pretrainingcited by 109
  12. 2025TerraMind: Large-Scale Generative Multimodality for Earth Observationcited by 104
  13. 2023Lightweight, Pre-trained Transformers for Remote Sensing Timeseriescited by 103
  14. 2021Learnable Fourier Features for Multi-Dimensional Spatial Positional Encodingcited by 93
  15. 2024HyperSIGMA: Hyperspectral Intelligence Comprehension Foundation Modelcited by 73
  16. 2024PANGAEA: A Global and Inclusive Benchmark for Geospatial Foundation Modelscited by 68
  17. 2024OmniSat: Self-Supervised Modality Fusion for Earth Observationcited by 63
  18. 2025Towards a Unified Copernicus Foundation Model for Earth Visioncited by 53
  19. 2024Bridging Remote Sensors with Multisensor Geospatial Foundation Modelscited by 46
  20. 2022Masked Image Modeling with Denoising Contrastcited by 45
  21. 2024SpectralEarth: Training Hyperspectral Foundation Models at Scalecited by 42
  22. 2025TESSERA: Temporal Embeddings of Surface Spectra for Earth Representation and Analysiscited by 39
  23. 2025Panopticon: Advancing Any-Sensor Foundation Models for Earth Observationcited by 34
  24. 2023FoMo: Multi-Modal, Multi-Scale and Multi-Task Remote Sensing Foundation Models for Forest Monitoringcited by 31
  25. 2023USat: A Unified Self-Supervised Encoder for Multi-Sensor Satellite Imagerycited by 28
  26. 2025SkySense V2: A Unified Foundation Model for Multi-modal Remote Sensingcited by 26
  27. 2025TerraFM: A Scalable Foundation Model for Unified Multisensor Earth Observationcited by 25
  28. 2025OlmoEarth: Stable Latent Image Modeling for Multimodal Earth Observationcited by 24
  29. 2024Towards Latent Masked Image Modeling for Self-Supervised Visual Representation Learningcited by 20
  30. 2024SenPa-MAE: Sensor Parameter Aware Masked Autoencoder for Multi-Satellite Self-Supervised Pretrainingcited by 18
  31. 2025EarthView: A Large Scale Remote Sensing Dataset for Self-Supervisioncited by 11
  32. 2024Planted: a dataset for planted forest identification from multi-satellite time seriescited by 9
  33. 2025SMARTIES: Spectrum-Aware Multi-Sensor Auto-Encoder for Remote Sensing Imagescited by 9
  34. 2025FLAIR-HUB: Large-scale Multimodal Dataset for Land Cover and Crop Mappingcited by 6
  35. 2022CroCo: Cross-Modal Contrastive learning for localization of Earth Observation datacited by 4
  36. 2025DUNIA: Pixel-Sized Embeddings via Cross-Modal Alignment for Earth Observation Applicationscited by 4
  37. 2025RAMEN: Resolution-Adjustable Multimodal Encoder for Earth Observationcited by 4
  38. 2026Any Model, Any Place, Any Time: Get Remote Sensing Foundation Model Embeddings On Demandcited by 4

50 most cited of 53 All 53 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026TESSERA v2: Scaling Pixel-wise Earth Foundation Modelscited by 5
  2. 2026Multi-Modal, Multi-Temporal Geospatial Foundation Models Improve Global Above-Ground Biomass Estimationcited by 1
  3. 2026Benchmarking Hyperspectral Foundation Models for Hyperspectral Unmixingcited by 0
  4. 2026Atomizer-IO: Beyond Pixels, Patches and Gridscited by 0
  5. 2026Autoregressive Drillhole Modelling Under Distribution Shiftcited by 0