Research map / Label-efficient and robust learning / Self-supervised and contrastive
Masked Angle-Aware Autoencoder for Remote Sensing Images
Zhihao Li, Biao Hou, Siteng Ma, zitong wu, Xianpeng Guo, bo ren, Licheng Jiao
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 |
|---|---|---|---|
| use scaling center crop instead of simple random rotation to create rotated crops (ablation) | MA3E · NWPU-RESISC45 fine-tuning | top-1 accuracy: 93.82 -> 95.77 | Improved |
| add OT loss for rotated crop reconstruction (with all other components) (ablation) | MA3E · DOTA1.0 | mAP50: 76.70 -> 77.93 | Improved |
| add OT loss on top of scaling center crop only (ablation) | MA3E · NWPU-RESISC45 fine-tuning | top-1 accuracy: 95.43 -> 95.06 | Hurt |
| add all MA3E components to MAE (ablation) | MAE · iSAID | mIoU: 60.96 -> 62.74 | Improved |
| add angle embedding to rotated crop (ablation) | MA3E · DOTA1.0 | mAP50: 76.12 -> 76.41 | Improved |
| increase rotated crop side length from 96 to 128 (ablation) | MA3E · DOTA1.0 | mAP50: 77.93 -> 76.92 | Hurt |
| use more non-overlapping rotated crops per image (ablation) | MA3E · NWPU-RESISC45 fine-tuning | top-1 accuracy: 95.21 -> 94.16 (a=32, 1 vs 4 crops) | Hurt |
| choose rotated crop position by selective search instead of random (ablation) | MA3E · DOTA1.0 | mAP50: 77.93 -> 78.08 | No clear effect |
| widen rotation range from 45 to 90 degrees (ablation) | MA3E · DOTA1.0 | mAP50: 77.93 -> 76.45 | Hurt |
| use block-wise masking instead of random masking (ablation) | MA3E · NWPU-RESISC45 fine-tuning | top-1 accuracy: 95.77 -> 94.98 | Hurt |
| pre-train longer (300 to 1600 epochs) | MA3E · DOTA1.0 | mAP50: 77.93 -> 79.47 | Improved |
Closest papers by meaning
- 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 0.85)
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.84)
- RoMA: Scaling up Mamba-based Foundation Models for Remote Sensing (NeurIPS 2025 · similarity 0.83)
- RSAR: Restricted State Angle Resolver and Rotated SAR Benchmark (CVPR 2025 · similarity 0.82)
- SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery (CVPR 2024 · similarity 0.82)
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.
