This is a simple, working example of "image segmentation" using a neural net trained by Jonathan Long and Evan Shelhamer, as described in Fully Convolutional Networks for Semantic Segmentation.
Trained model weights are from the Caffe Model Zoo.
Clone this repo, then:
# download weights from model zoo
CAFFE_ROOT=/path/to/caffe/repo make fcn-8s/fcn-8s-pascalcontext.caffemodel
# build docker container
make build-docker
If you're running docker host in a VM (i.e., on a Mac), make sure to forward port 8888.
Then do:
docker run -it --rm -v $(pwd):/workspace -p 8888:8888 caffe-fcn ./notebook.sh
Then go to http://localhost:8888