Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
On the Generalization of Representation Uncertainty in Earth Observation
Spyros Kondylatos, Nikolaos Ioannis Bountos, Dimitrios Michail, Xiao Xiang Zhu, Gustau Camps-Valls, Ioannis Papoutsis
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 the uncertainty module on EO data instead of ImageNet | ViT uncertainty module · EO inference datasets | LA-CPA / R-CPA | Improved |
| use ImageNet-pretrained uncertainty module on EO data | ViT uncertainty module · EO inference datasets | LA-CPA | Hurt |
| coarser pretraining targets (BigEarthNet-5 instead of BigEarthNet) (ablation) | ViT-Tiny and ViT-Small · EO inference datasets | LA-CPA | Improved |
| coarser pretraining targets (BigEarthNet-5 instead of BigEarthNet) (ablation) | ViT-Base and ViT-Large · EO inference datasets | LA-CPA | No clear effect |
| pretrain at original 120x120 resolution rather than downscaled (ablation) | ViT-Large on BigEarthNet · MLRSNet, Woody, Waititu, Flair | LA-CPA | Improved |
| downscale pretraining resolution to 60x60 or 30x30 (ablation) | ViT-Large on BigEarthNet · BigEarthNet, MARIDA | LA-CPA | No clear effect |
| downscale pretraining resolution to 16x16 (ablation) | ViT-Large on BigEarthNet · all datasets | LA-CPA | Hurt |
| pretrain on Flair rather than BigEarthNet | ViT uncertainty module · Woody and Waititu | LA-CPA | Improved |
| use max pixel loss instead of mean pixel loss in discard test (ablation) | EO-pretrained uncertainty · Flair discard test | downstream loss trend | Improved |
| EO-pretrained (Flair) vs ImageNet uncertainty on noisy BigEarthNet | ViT variants · noisy vs noise-free BigEarthNet | predicted uncertainty | Improved |
| use SAR instead of RGB input | BigEarthNet-pretrained ViT · BigEarthNet | LA@1 | Hurt |
| use SAR instead of RGB input | ViT-Base and ViT-Large · BigEarthNet | LA-CPA | Mixed |
Closest papers by meaning
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.87)
- REOBench: Benchmarking Robustness of Earth Observation Foundation Models (NeurIPS 2025 · similarity 0.86)
- See What We Cannot See: A Geo-guided Reasoning Benchmark for Object Counting under Adverse Earth Observation Conditions (CVPR 2026 · similarity 0.86)
- No Location Left Behind: Measuring and Improving the Fairness of Implicit Representations for Earth Data (ICLR 2025 · similarity 0.86)
- InstructSAM: A Training-free Framework for Instruction-Oriented Remote Sensing Object Recognition (NeurIPS 2025 · similarity 0.86)
- Towards Faithful Reasoning in Remote Sensing: A Perceptually-Grounded GeoSpatial Chain-of-Thought for Vision-Language Models (ICLR 2026 · similarity 0.86)
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.
