Research map / Detection and segmentation / Open-set and few-shot detection
Local Precise Refinement: A Dual-Gated Mixture-of-Experts for Enhancing Foundation Model Generalization against Spectral Shifts
Xi Chen, Maojun Zhang, Yu Liu, Shen Yan
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 mixture-of-experts (multiple experts instead of one) (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-sensor) | mIoU: 63.41 -> 66.19 | Improved |
| add mixture-of-experts (multiple experts instead of one) (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-regional) | mIoU: 59.37 -> 60.32 | Improved |
| add mixture-of-experts (multiple experts instead of one) (ablation) | SpectralMoE (DINOv3-Large) · FLAIR (cross-regional) | mIoU: 60.82 -> 63.18 | Improved |
| use separate gating networks for visual and depth (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-sensor) | mIoU: 63.44 -> 66.19 | Improved |
| use separate gating networks for visual and depth (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-regional) | mIoU: 58.97 -> 60.32 | Improved |
| use separate gating networks for visual and depth (ablation) | SpectralMoE (DINOv3-Large) · FLAIR (cross-regional) | mIoU: 61.95 -> 63.18 | Improved |
| add depth features from PromptDA (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-sensor) | mIoU: 63.52 -> 66.19 | Improved |
| add depth features from PromptDA (ablation) | SpectralMoE (DINOv3-Large) · FLAIR (cross-regional) | mIoU: 63.07 -> 63.18 | No clear effect |
| fuse depth via cross-attention instead of simple addition (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-regional) | mIoU: 58.35 -> 60.32 | Improved |
| fuse depth via cross-attention instead of simple addition (ablation) | SpectralMoE (DINOv3-Large) · Five-Billion-Pixels (cross-sensor) | mIoU: 63.48 -> 66.19 | Improved |
| increase number of experts beyond 6 (ablation) | SpectralMoE · FBP-CS/FBP-CR/FLAIR-CR | mIoU: peak at Ne=6 | Hurt |
| SpectralMoE adapter instead of REIN/DepthForge/SET on frozen backbone | DOFA/DINOv3 · 7 DGSS benchmarks | mIoU: e.g. 56.61 -> 59.83 (WHU-OHS) | Improved |
Closest papers by meaning
- CrossEarth: Geospatial Vision Foundation Model for Domain Generalizable Remote Sensing Semantic Segmentation (TPAMI 2025 · similarity 0.90)
- Earth-Adapter: Bridge the Geospatial Domain Gaps with a Frequency-Guided Mixture of Adapters (AAAI 2026 · similarity 0.89)
- Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation (AAAI 2025 · similarity 0.88)
- Stronger Fewer & Superior: Harnessing Vision Foundation Models for Domain Generalized Semantic Segmentation (CVPR 2024 · similarity 0.87)
- GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective (CVPR 2026 · similarity 0.87)
- Meta-Learning Hyperparameters for Parameter Efficient Fine-Tuning (CVPR 2025 · 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.
