Trained model (contained in model folder) where a pre-trained .h5 keras model (trained using 2180 images) is used for distinguishing between PPL and XPL thin sections
The model was trained on 2180 images (1090 XPL, 1090 PPL), due to a lack of computational power I reduced the image quality from around 1000x1000 px to 400x400px to train the model. All images are RGB.
The Training dataset and valitadion dataset were splitted 80% - 20% and data augmentation was used (width shift, height shift, zoom, flip).
Standard model for binary classification with data augmentation:
Maybe should have continued with more epochs...
-- UNAL, Iván Ferreira