Adapted from https://www.tensorflow.org/hub/tutorials/image_retraining
You can build a docker image to use by using make:
make
Or just don't use docker at all:
pip install tensorflow tensorflow-hub
docker run -v /your/train/images/:/data -v $(pwd)tf_files:/model/tf_files landmark /bin/bash /model/train.sh
You will get back trained models in your current folder (in tf_files/).
./train.sh /your/train/images/
docker run -v /your/test/images/:/data -v $(pwd):/result landmark /bin/bash /model/predict.sh
You will get your submission.csv
in your current folder.
./predict.sh /your/test/images/ submission.csv