Explanation of the approach and evaluation can be found in the report: urban_growth_report.pdf.

1. Clone the repo git clone [email protected]:tomas2211/urban_growth_task.git
2. Install requrements pip install -r requirements.txt
3. Download and unzip the dataset ./download_data.sh [link from task assignment]

If you download the dataset elsewhere, specify the path by --data_folder parameter. The dataset folder must contain images in imgs folder and labels in labels folder.

Pre-trained models can be found in models folder.

To visualize the urban index timeseries and save the figures in visualizations folder, use the following command:

python create_timeseries.py --device [cpu|cuda] --checkpoint_path models/[checkpoint] --out_folder visualizations

Training scripts with all parameter settings are located in scripts folder. Execute the script from the main directory.

Trained segmentation models can be evaluated by executing the following command:

python eval_net.py --device [cpu|cuda] --checkpoint_path models/[checkpoint] --out_folder evaluation_figures

If you want to enjoy the data analysis figures from the report as individual images, generate them using script data_analysis.py.