Research map / Vision-language and multimodal / Cross-modal learning
MMEarth: Exploring Multi-Modal Pretext Tasks For Geospatial Representation Learning
Vishal Nedungadi, Ankit Kariryaa, Stefan Oehmcke, Serge Belongie, Christian Igel, Nico Lang
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 |
|---|---|---|---|
| replace ImageNet pretraining data with Sentinel-2 RGB at patch size 32 (ablation) | ConvNeXt V2 MAE · BigEarth20k / SAcrop3k fine-tuning | F1 / IoU: SAcrop3k 26.7 -> 23.8; BigEarth20k FT 55.7 -> 55.5 | No clear effect |
| reduce patch size from 32 to 16 for 112px satellite images (ablation) | MAE on Sentinel-2 RGB · So2Sat20k fine-tuning | accuracy: 38.1 -> 48.8 | Improved |
| use all 12 Sentinel-2 bands instead of RGB only (ablation) | MAE on Sentinel-2 · So2Sat20k fine-tuning | accuracy: 54.1 -> 48.7 | Hurt |
| use all 12 Sentinel-2 bands instead of RGB only (ablation) | MAE on Sentinel-2 · SAcrop3k fine-tuning | IoU: 33.1 -> 36.0 | Improved |
| add multi-modal pretext tasks to Sentinel-2 reconstruction (ablation) | MP-MAE · BigEarth20k linear probing | F1: 36.2 -> 40.9 | Improved |
| add multi-modal pretext tasks to Sentinel-2 reconstruction (ablation) | MP-MAE · So2Sat20k linear probing | accuracy: 33.5 -> 41.7 | Improved |
| add only pixel-level or only image-level pretext tasks (ablation) | MP-MAE · BigEarth20k linear probing | F1: 36.2 -> 36.6 | No clear effect |
| combine pixel-level and image-level pretext groups (ablation) | MP-MAE · BigEarth20k linear probing | F1: 36.6 -> 40.9 | Improved |
| multi-modal pretext tasks in few-shot linear probing | MP-MAE · EuroSat2k 1% of labels | accuracy: 16.5% -> 34% | Improved |
| MP-MAE (3.7M encoder) versus ResNet-50 (25M) baselines on linear probing BigEarth20k | MP-MAE · BigEarth20k linear probing | F1: 44.5 -> 43.3 (vs SeCo) | Hurt |
Closest papers by meaning
- SkySense: A Multi-Modal Remote Sensing Foundation Model Towards Universal Interpretation for Earth Observation Imagery (CVPR 2024 · similarity 0.88)
- Galileo: Learning Global & Local Features of Many Remote Sensing Modalities (ICML 2025 · similarity 0.87)
- InstructSAM: A Training-free Framework for Instruction-Oriented Remote Sensing Object Recognition (NeurIPS 2025 · similarity 0.87)
- EarthDial: Turning Multi-sensory Earth Observations to Interactive Dialogues (CVPR 2025 · similarity 0.87)
- GeoCore-9B: Towards Geo-Aware Generative Foundation Models in Earth Observation (NeurIPS 2026 · similarity 0.87)
- GeoLLaVA-8K: Scaling Remote-Sensing Multimodal Large Language Models to 8K Resolution (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.
