Research map / Architectures and efficiency / Pruning and efficient tuning
RingMo-Aerial: An Aerial Remote Sensing Foundation Model With Affine Transformation Contrastive Learning
Wenhui Diao, Haichen Yu, Kootak Kang, Tong Ling, Di Liu, Yingchao Feng, Hanbo Bi, Libo Ren, Xuexue Li, Yongqiang Mao, Xian Sun
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 FE-MSA and MIM pre-training on aerial data (RingMo-Aerial MIM vs RingMo) (ablation) | RingMo-Aerial · UAVid segmentation (UperNet) | mIoU: 61.9 -> 65.8 | Improved |
| add affine-transformation contrastive learning after MIM pre-training (ablation) | RingMo-Aerial · UAVid segmentation (UperNet) | mIoU: 65.8 -> 66.4 | Improved |
| add affine-transformation contrastive learning after MIM pre-training (ablation) | RingMo-Aerial · VisDrone-DET (Cascade RCNN) | mAP: 30.5 -> 31.1 | Improved |
| add contrastive learning pre-training when fine-tuning with ARS-Adapter (ablation) | RingMo-Aerial + ARS-Adapter · VisDrone-DET | mAP: 25.3 -> 31.0 | Improved |
| add contrastive learning pre-training when fine-tuning with ARS-Adapter (ablation) | RingMo-Aerial + ARS-Adapter · UAVid | mIoU: 62.9 -> 64.4 | Improved |
| use ARS-Adapter instead of Adapter for parameter-efficient fine-tuning (ablation) | RingMo-Aerial · UAVid | mIoU: 57.7 -> 64.4 | Improved |
| use ARS-Adapter instead of AdaptFormer (ablation) | RingMo-Aerial · UAVid | mIoU: 62.1 -> 64.4 | Improved |
| use ARS-Adapter (0.41% params) instead of full fine-tuning (ablation) | RingMo-Aerial · VisDrone-DET | mAP: 31.1 -> 31.0 | No clear effect |
| raise pre-training mask ratio from 0.6 to 0.8 (ablation) | RingMo-Aerial · DroneVehicle RGB+IR | mAP_s: 0.228 -> 0.235 | Mixed |
| change mask ratio from 0.6 to 0.7 (ablation) | RingMo-Aerial · UAVid | mIoU: 68.70 -> 67.72 | Hurt |
| place depthwise conv before attention rather than after FFN (ablation) | FE-MSA · DroneVehicle RGB+IR | mAP: 0.485 -> 0.492 | Improved |
| use depthwise kernel size 7 instead of 3 (ablation) | FE-MSA · DroneVehicle RGB+IR | mAP: 0.485 -> 0.492 | 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.
- AdaptFormer — AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition (NeurIPS 2022) · better on 2 of 2 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
- ZoomEarth: Active Perception for Ultra-High-Resolution Geospatial Vision-Language Tasks (CVPR 2026 · similarity 0.82)
- RoMA: Scaling up Mamba-based Foundation Models for Remote Sensing (NeurIPS 2025 · similarity 0.81)
- FBRT-YOLO: Faster and Better for Real-Time Aerial Image Detection (AAAI 2025 · similarity 0.81)
- FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection (AAAI 2024 · similarity 0.80)
- CHOICE: Benchmarking the Remote Sensing Capabilities of Large Vision-Language Models (NeurIPS 2025 · similarity 0.80)
- Meta-Learning Hyperparameters for Parameter Efficient Fine-Tuning (CVPR 2025 · similarity 0.80)
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.
