Research map / Vision-language and multimodal / Human and embodied modeling
SpectralGPT: Spectral Remote Sensing Foundation Model
Danfeng Hong, Bing Zhang, Xuyang Li, Yuxuan Li, Chenyu Li, Jing Tao Yao, Naoto Yokoya, Hao Li, Pedram Ghamisi, Xiuping Jia, Antonio Plaza, Paolo E. Gamba and 2 more
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 SpectralGPT pretraining on top of random init (patch 8) (ablation) | ViT-Base · BigEarthNet | mAP: 80.15 -> 85.61 | Improved |
| use patch size 8 instead of 16 (ablation) | SpectralGPT · BigEarthNet | mAP: 76.42 -> 85.61 | Improved |
| pretrain on fMoW-S2 instead of BigEarthNet only (ablation) | SpectralGPT · BigEarthNet | mAP: 83.97 -> 85.61 | Improved |
| add BigEarthNet continuation pretraining after fMoW-S2 (ablation) | SpectralGPT · BigEarthNet | mAP: 85.61 -> 87.51 | Improved |
| raise masking ratio from 25% to 90% (ablation) | SpectralGPT · BigEarthNet | mAP: 82.73 -> 85.61 | Improved |
| raise masking ratio from 75% to 90% (ablation) | SpectralGPT · BigEarthNet | mAP: 84.64 -> 85.61 | Improved |
| normalize the reconstruction target (ablation) | SpectralGPT · BigEarthNet | mAP: 84.95 -> 85.61 | Improved |
| normalization instead of standardization (ablation) | SpectralGPT · BigEarthNet | mAP: 85.39 -> 85.61 | Improved |
| decoder depth 4 instead of 2 (ablation) | SpectralGPT · BigEarthNet | mAP: 84.59 -> 85.61 | Improved |
| increase decoder depth from 4 to 8 blocks (ablation) | SpectralGPT · BigEarthNet | mAP: 85.61 -> 85.38 | Hurt |
| scale ViT-Base to ViT-Huge | SpectralGPT · BigEarthNet | micro-mAP: 85.61 -> 89.23 | Improved |
| add BigEarthNet to pretraining data | SpectralGPT · SegMunich | mIoU: 49.8 -> 51.0 | Improved |
Closest papers by meaning
- S2MAE: A Spatial-Spectral Pretraining Foundation Model for Spectral Remote Sensing Data (CVPR 2024 · similarity 0.90)
- LESSViT: Robust Hyperspectral Representation Learning under Spectral Configuration Shift (NeurIPS 2026 · similarity 0.88)
- MMM-RS: A Multi-modal, Multi-GSD, Multi-scene Remote Sensing Dataset and Benchmark for Text-to-Image Generation (NeurIPS 2024 · similarity 0.87)
- Parameter Efficient Self-Supervised Geospatial Domain Adaptation (CVPR 2024 · similarity 0.87)
- RoMA: Scaling up Mamba-based Foundation Models for Remote Sensing (NeurIPS 2025 · similarity 0.86)
- Pan-LUT: Efficient Pan-sharpening via Learnable Look-Up Tables (NeurIPS 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.
