You will need to generate unit visualizations before using the web interface. Please use the code in unit_visualization to do so. You will also need to modify the static/raw symlink to point to the correct location of ddsm_raw if you are not using the default data locations.
Next, set up the static/unit_vis and static/unit_vis_subset directories. The static/unit_vis directory is already set up as a symlink to the output images generated by unit visualization. Since some CNN layers have many more units than is feasible for annotation, we also create symlinks in the static/unit_vis_subset directory to only show a subset of the units in each layer.
If you used our pretrained models to generate unit visualizations, the static/unit_vis and static/unit_vis_subset directories are already set up properly, and should exactly reproduce the visualizations shown to our expert annotators. If you would like to visualize your own trained models, you can modify the symlinks accordingly.
Please run the following script to cache the validation set patch labels into a .pickle file:
python cache_patch_labels.pyTo start up the server, use the following commands:
export FLASK_APP=server.py
flask run -h 0.0.0.0You can then visit localhost:5000 in a web browser to view the annotation web interface.
A summary of all annotations from all expert annotators is shown at localhost:5000/summary.