Research map / Label-efficient and robust learning / Self-supervised and contrastive
CROMA: Remote Sensing Representations with Contrastive Radar-Optical Masked Autoencoders
Anthony Fuller, Koreen Millard, James R Green
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 |
|---|---|---|---|
| combine contrastive and MAE objectives (vs MAE only) (ablation) | CROMA-B · BigEarthNet optical linear probe | mAP: -6.0 change when removing contrastive | Improved |
| combine contrastive and MAE objectives (vs contrast only) (ablation) | CROMA-B · BigEarthNet optical linear probe | mAP: -3.0 change when removing MAE | Improved |
| 2D-ALiBi + X-ALiBi instead of 2D-sinusoidal (ablation) | CROMA-B · segmentation avg | mIoU: -1.7 when using sinusoidal | Improved |
| 2D-ALiBi instead of PEG (ablation) | CROMA-B · segmentation avg | mIoU: -0.7 when using PEG | Improved |
| add X-ALiBi on top of 2D-ALiBi (ablation) | CROMA-B · avg across six evals | mAP/mIoU: -0.2 without X-ALiBi | Improved |
| 2D-ALiBi for extrapolation to 504x504 images (ablation) | CROMA-B · DW-Expert-120 | mIoU: 19.5 -> 58.8 | Improved |
| lower mask ratio (independent 50% vs 75%) (ablation) | CROMA-B · avg ablation | mIoU: +0.1 | No clear effect |
| multimodal MAE target instead of radar-only (ablation) | CROMA-B · segmentation avg | mIoU: -0.5 with radar-only | Improved |
| larger MAE decoder (depth 6, dim 768) (ablation) | CROMA-B · segmentation avg | mIoU: +0.1 | No clear effect |
| scale to 300 epochs pretraining (ablation) | CROMA-B · segmentation avg | mIoU: +1.0 | Improved |
| add radar-optical joint representation vs optical only | CROMA · BigEarthNet linear probe | mAP: +1.2 | Improved |
| fMoW-Sentinel finetuning vs SatMAE | CROMA-B · fMoW-Sentinel finetuning | Top-1: 57.20 -> 54.47 | Hurt |
Later papers that compared with this method
2 later accepted papers report CROMA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Brewing Stronger Features: Dual-Teacher Distillation for Multispectral Earth Observation (CVPR 2026 · better on 2 of 2 shared results)
- TerraFM: A Scalable Foundation Model for Unified Multisensor Earth Observation (ICLR 2026 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SatMAE — SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery (NeurIPS 2022) · better on 8 of 8 shared results
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
- Galileo: Learning Global & Local Features of Many Remote Sensing Modalities (ICML 2025 · similarity 0.87)
- Learning De-Biased Representations for Remote-Sensing Imagery (NeurIPS 2024 · similarity 0.86)
- DiffCLIP: Few-shot Language-driven Multimodal Classifier (AAAI 2025 · similarity 0.85)
- LDS2AE: Local Diffusion Shared-Specific Autoencoder for Multimodal Remote Sensing Image Classification with Arbitrary Missing Modalities (AAAI 2024 · similarity 0.85)
- SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery (CVPR 2024 · similarity 0.85)
- RobSense: A Robust Multi-modal Foundation Model for Remote Sensing with Static, Temporal, and Incomplete Data Adaptability (CVPR 2025 · similarity 0.85)
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.
