VGG 16 model

[ArturAzarskyy]

Try to use Jordys definition of VGG16 model in GrzegorzMeller solution:

link to the model : https://github.com/JordiCorbilla/ocular-disease-intelligent-recognition-deep-learning/blob/f67ed307c3963e8e06c676f708e448b6663b604c/odir_model_vgg16.py#L21

link to how he trains the model (we could just check if he does additional things there):

• vgg-total https://github.com/JordiCorbilla/ocular-disease-intelligent-recognition-deep-learning/blob/master/odir_vgg_training.py
• vvg-16 training https://github.com/JordiCorbilla/ocular-disease-intelligent-recognition-deep-learning/blob/master/odir_vgg16_training_basic.py

[ArturAzarskyy]

• There are many differences between two VGG models for GrzegorzMeller and Jordi's models
• Different Optimizers: Adam vs SGD
• Different stride size
• number of Conv2D layers per block after 3rd block increases

Notes on progress :

• From current cun with 3 layers it looks weird, low test accuracy but high validation accuracy ??
• I'll try to adapt how Jordi fully did his model
• might need to onsider to save the current weights to the file and preload the weights into the model after each run just like the jordi does