Get Started with Sample Code

🌍 OpenEarthMap Land Cover Mapping

📌 Starter Kit

🔹 Simple U-Net implementation with instructions for compiling the xBD dataset.

🔹 Ready-to-run Google Colaboratory demo of U-Net-EfficientNet-B4, based on the work from the segmentation_models.pytorch repository by qubvel, available at: https://github.com/qubvel/segmentation_models.pytorch.

📌 Lightweight Models

🔹 Demo of efficient OpenEarthMap models optimized with SparseMask and FasterSeg neural architecture search methods. Models are automatically searched & pretrained on OpenEarthMap (training & validation sets).

🏆 OpenEarthMap Few-Shot Challenge

🔹 Baseline model for few-shot semantic segmentation in land cover mapping.
🧑‍💻 Code Repository: GitHub

🗺️ Multimodal Change Detection

📌 ObjFormer

🔹 ObjFormer is a robust and efficient model for detecting changes from optical remote sensing imagery and map data. It combines object-based image analysis techniques with self-attention mechanisms to enhance accuracy and efficiency in change detection tasks.

🌍 3D Semantic Understanding

📌 RS3DAda

🔹 Multi-task unsupervised domain adaptation (UDA) & domain generalization (DG) framework for improving 3D semantic understanding from synthetic data.

🏆 OpenEarthMap-SAR Challenge

🔹 Baseline model for SAR-based land cover mapping in the 2025 IEEE GRSS DFC Track 1.
🧑‍💻 Code Repository: GitHub