This projects works with https://www.kaggle.com/c/dogs-vs-cats/data they have to be 2000 total and 1000 for each dogs and cats folder. They have to be in a main folder data and subfolders(validation/cat, validation/dogs and the same for train). In validation there have to be 400 extra photos.
1.fit_generator for training Keras a model using Python data generators 2.ImageDataGenerator for real-time data augmentation 3.layer freezing and model fine-tuning
ReLU stack of 3 convolution layers Yann LeCun advocated in the 1990s for image classification binary_crossentropy loss to train our model.
The method use does not repeat photos(shuffle: false) and create there copies which it rotates. The method use pretrained database (VGG16 network - ImageNet dataset).
The folder structure date/train/ for training data/validation/ for testing the algorithm
epochs - for one iteration images batch_size - all iterations