Code for our implementation of thorax X-Ray image classification and localization, as part of the Project Course in Data Science for replicating and improving the results of the paper ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases.
Four full information regarding the appraoches we took to tackle the problem and the results, please see our report.
There are a few arguments that can be used when running the program:
--evaluate
: If used, the chosen model will be used for evaluation, that is, obtaining ROC curves, AUC numbers, or generating bounding boxes.
--model_path
: The path to the model which is to be evaluated (only needed in the --evaluate
flag is used).
--use_comet
: If used, the experiment will be trakced in comet. Now only works for Moein's API key, but the experiment is public and everyone can view it.
--save_checkpoints
: If used, the model checkpoints will be saved (at every epoch).
--lr
: Determines the learning rate to be used.
--wdecay
: If set, weight decay is applied (see the code for more details).
--max_epochs
: determines the maximum number of epochs to train the model
--simple_lr_decay
: If used, a simple version of learning rate decay will be used (see the code for more details).
----net_type
: Determines which network type should be used for training. It could be unified_net
, attention1
, attention2
, or attentionSE
. See the code for more details.