🛰️ Earth AI Remote Sensing Models

🤝 This is not an officially supported Google product.

This project is not eligible for the Google Open Source Software Vulnerability Rewards Program.

Visit the Earth AI website and check the technical report for additional information about Earth AI and the different model families.

📘 Research Papers

• FM-ChangeNet: Learning Change through Pathwise Feature Transport (abstract, pdf)
• RSRCC: A Remote Sensing Regional Change Comprehension Benchmark Constructed via Retrieval-Augmented Best-of-N Ranking (abstract, pdf)
• GroundSet: A Cadastral-Grounded Dataset for Spatial Understanding with Vector Data (joint work with Roger Ferrod and Sylvain Lobry from Université Paris Cité), ECCV 2026 (abstract, pdf)
• RS-OVC: Open-Vocabulary Counting for Remote-Sensing Data - pdf, ICPR 2026 slides, code. ArXiv version: (abstract, pdf)
• Type-Aware Ranking of Urban Similarity from Aerial Imagery (abstract, pdf)
• A Recipe for Improving Remote Sensing VLM Zero Shot Generalization (abstract, pdf)
• On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration (abstract, pdf)
• Zero-Shot Multi-Spectral Learning: Reimagining a Generalist Multimodal Gemini 2.5 Model for Remote Sensing Applications (abstract, pdf, blog, colab)
• Enhancing Remote Sensing Representations Through Mixed-Modality Masked Autoencoding (abstract, pdf)

📊 Datasets

• RSRCC - The dataset is designed for semantic change understanding in remote sensing, pairing multi-temporal image evidence with natural language questions and answers.
• GroundSet - This dataset by Roger Ferrod and Sylvain Lobry from Université Paris Cité can be used to train and evaluate VLMs on a variety of Remote Sensing tasks.

💻 Code repository

Implementation of the Few-shot algorithm from the On-the-Fly OVD Adaptation with FLAME: Few-shot Localization via Active Marginal-Samples Exploration paper.

Remote Sensing Models on Vertex AI Model Garden

Vertex AI Model Garden hosts the Classification and Retrieval Model for Remote Sensing (the MaMMUT embeddings model) and the Object Detection for Remote Sensing (based on the OWL-ViT architecture). Deployment instructions will be provided when access is granted.

⚠️ Additional useful notebooks:

• Model deployment notebook
• VLM Batch prediction