McAntek/building-segmentation-poland-b3

Building Segmentation Poland — SegFormer B3

A building-segmentation model based on SegFormer B3, developed to identify building footprints in aerial imagery of Poland.

Model

• Architecture: SegFormer B3
• Task: Semantic segmentation of buildings
• Region of interest: Poland
• Files: Model weights, model configuration, preprocessing configuration, and training history

Repository contents

• best_model/ — trained model weights and configuration
• history.json — training history

Usage

The model is provided for experimentation and further development in aerial-image building segmentation. To run inference, use a compatible Hugging Face Transformers environment and the appropriate preprocessing pipeline for the model.

The expected input bands, normalization, tile size, and output-label mapping should be confirmed from the training and inference code before use.

Intended use

Potential applications include building-footprint extraction, geospatial analysis, and research on automated mapping from aerial imagery. Predictions should be validated before use in operational mapping or other consequential applications.