Research map / Detection and segmentation / Open-set and few-shot detection
ATRNet-STAR: A Large Dataset and Benchmark Toward Remote Sensing Object Recognition in the Wild
Yongxiang Liu, Weijie Li, Li Liu, Jie Zhou, Bowen Peng, Yafei Song, Xuying Xiong, Wei Yang, Tianpeng Liu, Zhen Liu, Xiang Li
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 |
|---|---|---|---|
| train on simple scenes, test on complex scenes (SOC-40 to EOC-Scene) | ConvNeXt · ATRNet-STAR classification | overall accuracy (%): 96.0 -> 16.5 | Hurt |
| train on 15 degree depression, test on other angles (SOC-40 to EOC-Depression) | ConvNeXt · ATRNet-STAR classification | overall accuracy (%): 96.0 -> 43.1 | Hurt |
| train on azimuth 0-60 degrees, test on others (SOC-40 to EOC-Azimuth) | SARATR-X · ATRNet-STAR classification | overall accuracy (%): 96.4 -> 26.4 | Hurt |
| use complex-image methods instead of magnitude methods for scene shift | HDANet / LDSF / MS-CVNet · EOC-Scene classification | overall accuracy (%): 16.5 -> 29.7 (LDSF); 33.6 (HDANet) | Improved |
| use complex-image methods under band and polarization shift | MS-CVNet / LDSF · EOC-Band, EOC-Polarization classification | overall accuracy (%): 89.2 -> 60.9 (LDSF) on EOC-Band; 84.6 -> 55.1 on EOC-Polarization | Hurt |
| add self-supervised SAR pre-training (SARATR-X vs HiViT) | HiViT · SOC-40 classification | overall accuracy (%): 86.8 -> 96.4 | Improved |
| use two-stage detector instead of single-stage under EOC | Faster RCNN / DiffDet4SAR / SARATR-X · ATRNet-STAR detection | mAP50: EOC-Scene 21.5 (YOLOv10) -> 37.2 (DiffDet4SAR) | Improved |
| single-stage CenterNet under depression and polarization shift | CenterNet · EOC-Depression, EOC-Polarization detection | mAP50: 22.3 (YOLOv10) -> 33.5 EOC-Depression; 59.2 -> 74.8 EOC-Polarization | Improved |
| reduce training set size on logarithmic scale | classification models · SOC-50 | accuracy | Hurt |
| pre-train on ATRNet-STAR SOC-40 before 5-shot fine-tuning | classification models · vehicle and aircraft SAR datasets (5-shot) | accuracy | Improved |
| pre-train on ATRNet-STAR before 5-shot fine-tuning on ship data | classification models · ship SAR dataset (5-shot) | accuracy | No clear effect |
Closest papers by meaning
- SARDet-100K: Towards Open-Source Benchmark and ToolKit for Large-Scale SAR Object Detection (NeurIPS 2024 · similarity 0.88)
- RSAR: Restricted State Angle Resolver and Rotated SAR Benchmark (CVPR 2025 · similarity 0.86)
- M3LEO: A Multi-Modal, Multi-Label Earth Observation Dataset Integrating Interferometric SAR and Multispectral Data (NeurIPS 2024 · similarity 0.84)
- FUSAR-GPT: A Spatiotemporal Feature-Embedded and Two-Stage Decoupled Visual Language Model for SAR Imagery (CVPR 2026 · similarity 0.84)
- Cross-Rejective Open-Set SAR Image Registration (CVPR 2025 · similarity 0.83)
- Rethinking Transformers Pre-training for Multi-Spectral Satellite 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.
