This is Kelwin's code to his paper: https://ieeexplore.ieee.org/document/8489527
My changes (Ricardo):
- used the command-line
2to3
to port from Python2 to Python3 - modified the imports to use
tensorflow.compat.v1
API (compatibility API) - some minor changes to make it work: such as using
multiprocessing=True
since I was getting erros due to thread unsafe generator. - added a symlink:
data/partitions
to/data/bioeng/ordinal-segmentation/partitions
(adjust to your datapath) - I also added a baseline that uses 'softmax' instead of the 'sigmoid' (default)
- There are some warnings: I am ignoring them
How to use?
- Install latest TensorFlow:
pip3 install tensorflow
- Just run
python3 eval_ordinal_networks.py --dataset /data/bioeng/ordinal-segmentation/teeth.json
for theteeth
dataset. It will automatically train all models and evaluate them. - This will run for the first fold only. To run for other folds, you need to use
--first
and--last
.