Research map / Label-efficient and robust learning / Self-supervised and contrastive
TerraFM: A Scalable Foundation Model for Unified Multisensor Earth Observation
Muhammad Sohail Danish, Muhammad Akhtar Munir, Syed Roshaan Ali Shah, Muhammad Haris Khan, Rao Muhammad Anwer, Jorma Laaksonen, Fahad Shahbaz Khan, Salman Khan
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 modality as augmentation (ablation) | TerraFM-B · m-EuroSat/m-BigEarthNet | kNN accuracy: 83.20 -> 87.70 (EuroSat); 54.62 -> 57.63 (BigEarthNet) | Improved |
| add cross-attention fusion (ablation) | TerraFM-B · m-Cashew-Plant | mIoU: 59.17 -> 62.40 | Improved |
| add cross-attention fusion (ablation) | TerraFM-B · m-BigEarthNet | kNN F1: 57.63 -> 57.74 | No clear effect |
| add dual centering (ablation) | TerraFM-B · m-EuroSat/m-Cashew-Plant | kNN accuracy / mIoU: 88.50 -> 90.40 (EuroSat); 62.40 -> 64.58 (Cashew) | Improved |
| use global cross-attention with 196 queries instead of Q=5 spatial fusion (ablation) | TerraFM-B · m-EuroSat | kNN accuracy: 90.4 -> 77.1 | Hurt |
| replace convolutional patch embedding with ViT-S token extraction (ablation) | TerraFM-B · m-EuroSat | kNN accuracy: 90.4 -> 87.2 | Hurt |
| use more spatial queries (Q=5 vs Q=1) (ablation) | TerraFM-B · m-EuroSat | kNN accuracy: 89.2 -> 90.4 | Improved |
| use TerraFM instead of Galileo/CROMA baselines | TerraFM · Copernicus-Bench DFC2020-S2 | mIoU: CROMA 66.5 vs TerraFM 63.8 | Hurt |
| scale TerraFM-B to TerraFM-L | TerraFM · m-Brick-Kiln kNN | accuracy: 94.5 -> 93.0 | Hurt |
| use TerraFM-120M instead of Prithvi-EO-2.0 300M | TerraFM · Landslide4Sense | mIoU: 65.0 -> 70.8 | Improved |
Later papers that compared with this method
2 later accepted papers report TerraFM as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- GeoSANE: Learning Geospatial Representations from Models, Not Data (CVPR 2026 · better on 2 of 2 shared results)
- Brewing Stronger Features: Dual-Teacher Distillation for Multispectral Earth Observation (CVPR 2026 · better on 2 of 2 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- CROMA — CROMA: Remote Sensing Representations with Contrastive Radar-Optical Masked Autoencoders (NeurIPS 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery (CVPR 2024 · similarity 0.90)
- Galileo: Learning Global & Local Features of Many Remote Sensing Modalities (ICML 2025 · similarity 0.88)
- Parameter Efficient Self-Supervised Geospatial Domain Adaptation (CVPR 2024 · similarity 0.86)
- Bridging Remote Sensors with Multisensor Geospatial Foundation Models (CVPR 2024 · similarity 0.86)
- GeoLink: Empowering Remote Sensing Foundation Model with OpenStreetMap Data (NeurIPS 2025 · similarity 0.86)
- Terra: A Multimodal Spatio-Temporal Dataset Spanning the Earth (NeurIPS 2024 · 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.
