Irradiated/Irradiated_Calibrate/Iradiated_FL_TOPO_Meta 

• Performance metrics for all irradiated models in file starting with “all_notebooks” 

• Defect information within each model folder in csv files ending in “defects” and “defects_truth”, with “truth” for predicted images 

• Ripley results within folder “smallbayessegnet_lr1e-04_irradiated_Adam_new_augmentation_EWCE_5 “with “pred_true” folder for predicted images and “pred_false” for expert-labeled images

Unirradiated/Unirradiated_Calibrate/Uniradiated_FL_TOPO

• Performance metrics for all unirradiated models in file starting with “all_notebooks”
• Defect information within each model folder in csv files ending in “defects” and “defects_truth”, with “truth” for predicted images
• Ripley results within folder “segnet_lr1e-04_unirradiated_Adam_new_augmentation_EWCE_5” with “pred_true” folder for predicted images and “pred_false” for expert-labeled images

Irradiated_images/Unirradiated_images: All images used for training and testing

small_adam_irradiated_segnet_unirradiated_ewce.csv:

• Performance metrics per class copied from the performance_metrics_short.csv files within the folders for the best models for unirradiated and irraidated images(segnet_lr1e-04_unirradiated_Adam_new_augmentation_EWCE_5 and smallbayessegnet_lr1e-04_irradiated_Adam_new_augmentation_EWCE_5)

The models are Bayesian SegNest (BayesSegNest), Bayesian SegNet (Bayes), SegNet, and Small Bayesian SegNet (SmallBayes) models and using either a Expert Weighted Cross Entropy (EWCE) or Weighted Cross-Entropy (WCE) loss and Adam and RMS optimizer for the initial models. The models used for the additional adjustments are a SegNet model with a Adam optimizer and Focal (FL) or Topological (Topo) loss or a SegNet model with metadata and a Adam optimizer and EWCE loss