tdardinier / CIL

1. Setting up the data

The data has to be put in the data folder.

• The test images in data/test_images
• The training data in data/training/images and data/training/groundtruth

Then run python save_images.py to save the data in the relevant folders in .npy format.

2. Augment the data with morphological features

Once the data is set up (1): Run python morphological_feature_augmentation/feature_augmentation.py

3. Get the Kaggle submission

The data has to have been augmented (1 and 2).

1. Run python model_morpho_patch/train.py
2. Run python model_morpho_patch/classify.py
3. Run python simple_post_processing/post_processing.py

The submission should then be located at generated/submissions/morpho_patch_post/submission.csv

4. Get the comparison results

The baselines are all implemented in the pixelwise folder. The evaluation is also implemented in the pixelwise folder.

The data has to have been augmented (1 and 2). If you want to get the comparison data from our novel approach, run python python model_patch/crossval.py.

Then simply run python pixelwise/main.py. This will train the baselines on the subsets defined the 5-fold cross-validation, generate the predictions of the baselines, and use the predictions from the baselines (and from our novel approach) to show a table of summarized results similar (and with similar values) to the one in the report.