Research map / Label-efficient and robust learning / Self-supervised and contrastive
NeighborMAE: Exploiting Spatial Dependencies between Neighboring Earth Observation Images in Masked Autoencoders Pretraining
Liang Zeng, Valerio Marsocci, Wufan Zhao, Andrea Nascetti, Maarten Vergauwen
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 |
|---|---|---|---|
| pretrain with NeighborMAE instead of MAE | MAE ViT-L pretrained on fMoW · fMoW classification (linear probing) | accuracy: 66.8 -> 68.8 | Improved |
| pretrain with NeighborMAE instead of MAE | MAE ViT-L pretrained on fMoW · Five-Billion-Pixels segmentation (fine-tuned) | mIoU: +2.7% mIoU | Improved |
| enlarge input area instead of using neighbors (ablation) | MAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 58.0 -> 58.2 | No clear effect |
| sample uni-temporal neighboring images instead of base input (ablation) | MAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 58.0 -> 58.7 | Improved |
| use multi-temporal neighbors instead of uni-temporal neighbors (ablation) | NeighborMAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 58.7 -> 61.7 | Improved |
| use multi-temporal neighbors instead of uni-temporal neighbors (ablation) | NeighborMAE ViT-B pretrained on Satellogic · fMoW classification (frozen) | accuracy: 50.1 -> 52.4 | Improved |
| use co-located multi-temporal images instead of neighbors (ablation) | NeighborMAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 58.7 -> 58.2 | No clear effect |
| use dynamic mask ratio (0.75 to 0.85) instead of constant 0.75 (ablation) | NeighborMAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 61.0 -> 61.7 | Improved |
| raise upper bound of mask ratio from 0.85 to 0.90 (ablation) | NeighborMAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 61.7 -> 61.4 | Hurt |
| use input-visibility loss weighting instead of full reconstruction loss (ablation) | NeighborMAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 58.6 -> 61.7 | Improved |
| use visibility-based loss weight for cross-visible pixels instead of weight 1 (ablation) | NeighborMAE ViT-B pretrained on fMoW · fMoW classification (frozen) | accuracy: 61.6 -> 61.7 | No clear effect |
| use visibility-based loss weight for cross-visible pixels instead of weight 1 (ablation) | NeighborMAE ViT-B pretrained on Satellogic · fMoW classification (frozen) | accuracy: 51.8 -> 52.4 | Improved |
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.
- ScaleMAE — Scale-MAE: A Scale-Aware Masked Autoencoder for Multiscale Geospatial Representation Learning (ICCV 2023) · better on 1 of 1 shared result
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
- Decomposing Earth Embeddings with Sparse Autoencoders (NeurIPS 2026 · similarity 0.87)
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.85)
- Accounting for Spatial Variability with the Histogram of Oriented Gradients Based Masking Improves Performance of Masked Autoencoder over Hyperspectral Satellite Imagery (Student Abstract) (AAAI 2025 · similarity 0.85)
- S2MAE: A Spatial-Spectral Pretraining Foundation Model for Spectral Remote Sensing Data (CVPR 2024 · similarity 0.84)
- No Location Left Behind: Measuring and Improving the Fairness of Implicit Representations for Earth Data (ICLR 2025 · similarity 0.84)
- SatSynth: Augmenting Image-Mask Pairs through Diffusion Models for Aerial Semantic Segmentation (CVPR 2024 · similarity 0.83)
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.
