Code for master thesis

Map creation through semantic segmentation of aerial images using deep convolutional networks

Train deep convolutional neural networks on aerial images and learn them to create maps.

Everything is built with Keras and Tensorflow backend. We have code for two networks:

Name		Name	Last commit message	Last commit date
Latest commit History 823 Commits
.idea		.idea
docker		docker
keras_utils		keras_utils
networks		networks
preprocessing		preprocessing
.gitignore		.gitignore
Datasetstats.ipynb		Datasetstats.ipynb
Makefile		Makefile
README.md		README.md
Result plotting.ipynb		Result plotting.ipynb
Untitled.ipynb		Untitled.ipynb
augment.py		augment.py
datasetstats.py		datasetstats.py
lrtest_densenet.py		lrtest_densenet.py
lrtest_unet.py		lrtest_unet.py
make_dataset.py		make_dataset.py
map_generator.py		map_generator.py
map_generator_multiclass.py		map_generator_multiclass.py
meanstd.py		meanstd.py
merge_predictions.py		merge_predictions.py
new_pred.py		new_pred.py
predict.py		predict.py
predict_binary.py		predict_binary.py
predict_experiments.py		predict_experiments.py
predict_folder.py		predict_folder.py
predict_full.py		predict_full.py
requirements.txt		requirements.txt
run_densenet.py		run_densenet.py
run_finetune.py		run_finetune.py
run_unet.py		run_unet.py
run_unet_experiments.py		run_unet_experiments.py