Research map / Label-efficient and robust learning / Self-supervised and contrastive
SARMAE: Masked Autoencoder for SAR Representation Learning
Danxu Liu, Di Wang, Hebaixu Wang, Haoyang Chen, Wentao Jiang, Yilin Cheng, Haonan Guo, Wei Cui, Jing Zhang
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 MAE on SAR-1M instead of ImageNet-1K (ablation) | MAE ViT-B · FUSAR-SHIP | accuracy (%): 75.40 -> 82.22 | Improved |
| pretrain MAE on SAR-1M instead of ImageNet-1K (ablation) | MAE ViT-B · SSDD | mAP (%): 64.00 -> 64.20 | No clear effect |
| pretrain MAE on SAR-1M instead of ImageNet-1K (ablation) | MAE ViT-B · AIR-PolSAR-Seg | mIoU (%): 60.28 -> 64.36 | Improved |
| add SARE speckle-noise denoising to SAR-1M MAE (ablation) | MAE* ViT-B · FUSAR-SHIP | accuracy (%): 82.22 -> 86.80 | Improved |
| add SARE speckle-noise denoising to SAR-1M MAE (ablation) | MAE* ViT-B · AIR-PolSAR-Seg | mIoU (%): 64.36 -> 65.15 | Improved |
| add SARE speckle-noise denoising to SAR-1M MAE (ablation) | MAE* ViT-B · SSDD | mAP (%): 64.20 -> 64.40 | No clear effect |
| add SARC alignment to paired optical features (frozen DINOv3) (ablation) | MAE* (Add noise) ViT-B · SSDD | mAP (%): 64.4 -> 68.1 | Improved |
| add SARC alignment to paired optical features (ablation) | MAE* (Add noise) ViT-B · FUSAR-SHIP | accuracy (%): 86.80 -> 89.30 | Improved |
| add SARC alignment to paired optical features (ablation) | MAE* (Add noise) ViT-B · AIR-PolSAR-Seg | mIoU (%): 65.15 -> 66.53 | Improved |
| directly finetune DINOv3 on SAR instead of SARC alignment (ablation) | DINOv3 · FUSAR-SHIP | accuracy (%): 74.25 vs 89.30 (SARMAE) | Hurt |
| scale backbone from ViT-B to ViT-L | SARMAE · RSAR | mAP (%): 66.80 -> 72.20 | Improved |
| scale backbone from ViT-B to ViT-L | SARMAE · FUSAR-SHIP (30% labels) | accuracy (%): 92.92 -> 92.80 | No clear effect |
Closest papers by meaning
- SARDet-100K: Towards Open-Source Benchmark and ToolKit for Large-Scale SAR Object Detection (NeurIPS 2024 · similarity 0.85)
- FUSAR-GPT: A Spatiotemporal Feature-Embedded and Two-Stage Decoupled Visual Language Model for SAR Imagery (CVPR 2026 · similarity 0.83)
- RobSense: A Robust Multi-modal Foundation Model for Remote Sensing with Static, Temporal, and Incomplete Data Adaptability (CVPR 2025 · similarity 0.83)
- RSAR: Restricted State Angle Resolver and Rotated SAR Benchmark (CVPR 2025 · similarity 0.83)
- 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 0.82)
- Dual Information Purification for Lightweight SAR Object Detection (AAAI 2025 · 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.
