Research map / Label-efficient and robust learning / Self-supervised and contrastive
Scale-MAE: A Scale-Aware Masked Autoencoder for Multiscale Geospatial Representation Learning
Colorado J Reed, Ritwik Gupta, Shufan Li, Sarah Brockman, Christopher Funk, Brian Clipp, Kurt Keutzer, Salvatore Candido, Matt Uyttendaele, Trevor Darrell
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add GSD positional encoding to vanilla MAE (ablation) | Vanilla MAE · RESISC-45 kNN at 50% GSD | kNN accuracy: 72.8 -> 75.4 | Improved |
| add progressive Laplacian decoder to vanilla MAE (ablation) | Vanilla MAE · RESISC-45 kNN at 50% GSD | kNN accuracy: 72.8 -> 75.3 | Improved |
| combine GSDPE and Laplacian decoder (ablation) | Vanilla MAE · RESISC-45 kNN at 50% GSD | kNN accuracy: 72.8 -> 78.1 | Improved |
| reduce decoder layers from 8 to 3 (ablation) | Scale-MAE · RESISC-45 kNN at 100% GSD | kNN accuracy: 78.9 -> 80.7 | Improved |
| use 75% mask rate instead of 70% or 80% (ablation) | Scale-MAE · RESISC-45 kNN at 100% GSD | kNN accuracy: 79.3 / 79.9 -> 80.7 | Improved |
| reconstruct both low and high frequency separately instead of low only (ablation) | Scale-MAE · RESISC-45 kNN at 50% GSD | kNN accuracy: 72.9 -> 78.1 | Improved |
| reconstruct separate low/high targets instead of a combined image (ablation) | Scale-MAE · RESISC-45 kNN at 100% GSD | kNN accuracy: 80.2 -> 80.7 | Improved |
| Scale-MAE versus SatMAE on kNN across eight datasets | Scale-MAE · 8 remote sensing datasets kNN | average kNN accuracy: +5.6% | Improved |
| Scale-MAE versus ConvMAE on EuroSAT kNN | Scale-MAE · EuroSAT kNN | accuracy (%): 88.8 -> 86.7 | Hurt |
| Scale-MAE versus others on SpaceNet v1 at coarser GSD | Scale-MAE · SpaceNet v1 downscaled | mIoU gap: +0.9 mIoU at 100%, +1.2 at 75%, +1.7 at 50% | Improved |
| Scale-MAE decoder uses 3 layers instead of 8 | Scale-MAE · ViT-Large parameters | parameters: 329.5M -> 322.9M | Improved |
Later papers that compared with this method
4 later accepted papers report ScaleMAE as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- NeighborMAE: Exploiting Spatial Dependencies between Neighboring Earth Observation Images in Masked Autoencoders Pretraining (CVPR 2026 · better on 1 of 1 shared result)
- Brewing Stronger Features: Dual-Teacher Distillation for Multispectral Earth Observation (CVPR 2026 · better on 0 of 1 shared result)
- Learning Representations of Satellite Images From Metadata Supervision (ECCV 2024 · better on 1 of 1 shared result)
- Cross-Scale MAE: A Tale of Multiscale Exploitation in Remote Sensing (NeurIPS 2023 · better on 8 of 10 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SatMAE — SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery (NeurIPS 2022) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Cross-Scale MAE: A Tale of Multiscale Exploitation in Remote Sensing (NeurIPS 2023 · similarity 0.91)
- MMEarth: Exploring Multi-Modal Pretext Tasks For Geospatial Representation Learning (ECCV 2024 · similarity 0.89)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.85)
- R-MAE: Regions Meet Masked Autoencoders (ICLR 2024 · similarity 0.85)
- Rethinking Transformers Pre-training for Multi-Spectral Satellite Imagery (CVPR 2024 · similarity 0.85)
- Galileo: Learning Global & Local Features of Many Remote Sensing Modalities (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
