Research map / Label-efficient and robust learning / Self-supervised and contrastive
Disentangling the Benefits of Self-Supervised Learning to Deployment-Driven Downstream Tasks of Satellite Images (Student Abstract)
Zhuo Deng, Yibing Wei, Mingye Zhu, Xueliang Wang, Junchi Zhou, Zhicheng Yang, Hang Zhou, Zhenjie Cao, Lan Ma, Mei Han, Jui-Hsin Lai
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 SimCLR pretraining on BEN3 instead of ImageNet RGB pretraining | ResNet50 · NaSC-TG2, Train/Validation ratios 2:8, 5:5, and 8:2 | Top-1 accuracy: 92.85 -> 92.25; 93.64 -> 93.37; 94.05 -> 94.45 | Mixed |
| Use four spectral channels (BEN4, RGBN) rather than three (BEN3, RGB) for SimCLR pretraining | ResNet50 on NaSC-TG2 · NaSC-TG2, Train/Validation ratios 2:8, 5:5, and 8:2 | Top-1 accuracy: 92.25 -> 92.90; 93.37 -> 95.59; 94.45 -> 97.13 | Improved |
| Use MAE+ViT-B with BEN4 | NaSC-TG2 scene classification · NaSC-TG2, Train/Validation ratios 2:8, 5:5, and 8:2 | Top-1 accuracy: 93.28; 96.11; 98.30 | Improved |
| Use SimCLR-pretrained ResNet50 rather than ResNet50 without SSL | Potsdam segmentation · Potsdam | mIoU: 80.80 -> 82.76 | Improved |
| Use SimCLR-pretrained ResNet50 rather than ResNet50 without SSL | Potsdam segmentation · Potsdam | OA: 89.80 -> 91.01 | Improved |
| Add MAE pretraining on BEN4 | ViT-B+UperNet on Potsdam · Potsdam | mIoU: 76.57 -> 78.35 | Improved |
| Add MAE pretraining on BEN4 | ViT-B+UperNet on Potsdam · Potsdam | OA: 87.33 -> 88.41 | Improved |
| Use vanilla ViT-B rather than the CNN-based ResNet50 model | Potsdam segmentation without SSL · Potsdam | mIoU: 80.80 -> 76.57 | Hurt |
Closest papers by meaning
- When Does Self-supervision Improve Few-shot Learning? (ECCV 2020 · similarity 0.87)
- Change-Aware Sampling and Contrastive Learning for Satellite Images (CVPR 2023 · similarity 0.87)
- A Closer Look at Benchmarking Self-supervised Pre-training with Image Classification (IJCV 2025 · similarity 0.86)
- Learning Representations of Satellite Images From Metadata Supervision (ECCV 2024 · similarity 0.84)
- A Survey on Self-Supervised Learning: Algorithms, Applications, and Future Trends (TPAMI 2024 · similarity 0.84)
- Measuring Self-Supervised Representation Quality for Downstream Classification Using Discriminative Features (AAAI 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.
