VIDA-NYU / GDPFinder

GDP Finder

Granularity at Scale

Paper: Granularity at Scale: Estimating Neighborhood Socioeconomic Indicators from High-Resolution Orthographic Imagery and Hybrid Learning

For the supervised approach:

notebooks/supervised_approach.ipynb contains data (census and satellite imagery) preparation and analysis. Cells to test saved models and interpret results are also there.

Example usage to train a model:

Within scripts directory:

$ nohup python -u supervised_training.py --metric 'density' --imagetype 'resize' --newwidth 1234 --newheight 1234 &

See supervised_training.py for more details on the arguments and training process.

In supervised_approach.ipynb and supervised_training.py, the dataset is generated from scripts/create_dataset.py and the model from scripts/supervised_models.py