FLAIR-2¶
Modular training and evaluation framework for semantic segmentation of aerial and Sentinel-2 satellite imagery.
Overview¶
This project provides three training pipelines for the FLAIR-2 dataset:
- Aerial pipeline — high-resolution aerial imagery segmentation
- Sentinel pipeline — Sentinel-2 temporal sequence segmentation
- Multimodal pipeline — late fusion of aerial and Sentinel-2 models
Quick Start¶
# Install dependencies
uv sync
# Run aerial training
uv run python -m src.pipeline.pipeline --config configs/config.yaml
# Run Sentinel-2 training
uv run python -m src.pipeline.sentinel_pipeline --config configs/config_sentinel_tsvit.yaml
API Documentation¶
Browse the full API reference in the sidebar to explore all model architectures, training utilities, and loss functions.