DeepGlobe Road Extraction
Sep 2025
Finished
Semantic segmentation project for road extraction from satellite imagery using the DeepGlobe Road Extraction dataset. Implements and compares two model architectures: (1) UNet-based segmentation model from segmentation_models_pytorch with various backbones, (2) SegFormer (nvidia/segformer-b0-finetuned-ade-512-512), a ViT-based semantic segmentation model from HuggingFace Transformers. The full pipeline includes dataset preprocessing, model training with configurable hyperparameters, and testing/evaluation. All experiments and results are documented in a comprehensive Jupyter notebook that serves as the project report.
Technologies
AI
HuggingFace
Jupyter
Matplot
NumPy
Pandas
Plotly
Python
PyTorch
Sklearn
