Uses the VGG16 and VGG19 nets from the modelzoo. Minor changes in the *.prototxt
to adapt it to the new caffe version.
See net.ipynb.
# Download models
wget --show-progress -P 'model/' 'http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_16_layers.caffemodel'
wget --show-progress -P 'model/' 'http://www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_19_layers.caffemodel'
# Convert models to new caffe format (caffe needed)
# (Not strictly needed. Removes deprecation notice)
upgrade_net_proto_binary MODEL.caffemodel MODEL.new.caffemodel
# Build docker image
docker build docker/ -t caffe-jupyter
# Start jupyter notebook
./docker/start-jupyter.sh
# Open browser at 127.0.0.1:8888
# Enter 'a' as the password to enter the jupyter env
# Maybe you have to adapt some paths to the models. See net.ipynb
# Docker terminal to image to execute "upgrade_net_proto_binary" etc. and run python files normally
docker run --entrypoint --rm -v $(pwd):$(pwd) -w $(pwd) -it caffe-jupyter bash