🌍 Qartezator: Yet another aerial image-to-map translator

Qartezator is your translator between aerial images and maps.

Environment setup

Clone the repo: git clone https://github.com/AndranikSargsyan/qartezator.git

Set up virtualenv:

cd qartezator
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt 

If you need torch+cuda, you can use the following command

pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 --extra-index-url https://download.pytorch.org/whl/cu116

Demo

Download all models from this link and place inside models/ directory.

StreamLit

Start StreamLit demo by running:

streamlit run demo.py

CLI Inference

python -m qartezator.inference -m PATH-TO-MODEL -i PATH-TO-IMAGE -o OUTPUT-PATH

Training

Download training data from here and extract into data/ directory.

To start the training run

python -m qartezator.train --config-path ./qartezator/configs/qartezator-fourier.yaml

Evaluation

Download test images from here.

To do the inference on test set, run

python scripts/predict_many.py --source-dir SOURCE_IMG_DIR --model-path TRACED_MODEL_PATH --output-dir OUTPUT_DIR

please see more argument options in the script.

To evaluate PSNR, SSIM and L1, use

 python scripts/cal_psnr_ssim_l1.py --gt-path TARGET_MAPS_DIR --pred-path PREDICTED_MAPS_DIR

To calculate FID use

python -m pytorch_fid PREDICTED_MAPS_PATH TARGET_MAPS_PATH

Results

Qartezator-Fourier

Aerial image	Target map	Predicted map

Acknowledgements

Our work borrows code from the following repos:

https://github.com/advimman/lama

https://github.com/fenglinglwb/MAT