Research map / Label-efficient and robust learning / Domain adaptation and generalization
Get Your Embedding Space in Order: Domain-Adaptive Regression for Forest Monitoring
Sizhuo Li, Dimitri Gominski, Martin Brandt, Xiaoye Tong, Philippe Ciais
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 manifold diffusion for regression (MDR) on top of GOL embeddings (ablation) | GOL kNN, ViT-B/16 · DRIFT canopy height 5-shot avg | R2: 0.49 -> 0.59 | Improved |
| Add MDR without ordered (GOL) embeddings (ablation) | baseline kNN embeddings, ViT-B/16 · DRIFT canopy height avg | R2: 0.48 -> 0.40 | Hurt |
| Add geometric order learning (GOL) embedding regularization (ablation) | baseline kNN, ViT-B/16 · DRIFT canopy height avg | R2: 0.48 -> 0.49 | Improved |
| Add MDR to GOL on tree cover (ablation) | GOL, ViT-B/16 · DRIFT tree cover avg | R2: 0.72 -> 0.71 | No clear effect |
| Train with GOL instead of plain regression in source domain (ablation) | ViT-B/16 · Source domain test canopy height | R2: 0.88 -> 0.86 | No clear effect |
| Use GOL+MDR instead of GOL kNN for hard domain shift | Denmark to Spain canopy height · DRIFT 5-shot | R2: 0.21 -> 0.60 | Improved |
| Add linear calibration with support set to off-the-shelf regression | Regression, ViT-B/16 · DRIFT canopy height avg | R2: 0.23 -> 0.50 | Improved |
| Add fine-tuning with support set to GOL+MDR (ablation) | GOL+MDR, ViT-B/16 · DRIFT canopy height avg | R2: 0.59 -> 0.62 | Improved |
| Add fine-tuning to GOL+MDR on tree count (ablation) | GOL+MDR, ViT-B/16 · DRIFT tree count avg | R2: 0.56 -> 0.55 | No clear effect |
| Use transductive methods when the domain gap is large rather than inductive | DRIFT · DRIFT pairwise comparisons | R2 | Improved |
| Switch backbone from ViT-B/16 to VGG16 for fine-tuning (ablation) | FT, tree count · DRIFT 5-shot | R2: drop of 0.27 vs 0.03 for GOL+MDR | Hurt |
| Increase number of shots per group beyond 20 (ablation) | GOL+MDR · DRIFT height and count | R2 | No clear effect |
Closest papers by meaning
- FoMo: Multi-Modal, Multi-Scale and Multi-Task Remote Sensing Foundation Models for Forest Monitoring (AAAI 2025 · similarity 0.86)
- DUNIA: Pixel-Sized Embeddings via Cross-Modal Alignment for Earth Observation Applications (ICML 2025 · similarity 0.84)
- DiffusionSat: A Generative Foundation Model for Satellite Imagery (ICLR 2024 · similarity 0.84)
- AnySat: One Earth Observation Model for Many Resolutions, Scales, and Modalities (CVPR 2025 · similarity 0.84)
- Parameter Efficient Self-Supervised Geospatial Domain Adaptation (CVPR 2024 · similarity 0.84)
- Multi-modal Learning for Geospatial Vegetation Forecasting (CVPR 2024 · similarity 0.84)
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.
