Cats_vs_Dogs

Architecture

The architecture of the model has 2 blocks for feature extraction in which each has :

1. Convolution Layer
2. BatchNormalizing Layer
3. Pooling Layer
4. Dropout Layer

Following which is a 64 Neuron Fully Connected layer with 2 output Neurons

Compilation

For complilation :-

1. Optimizer = Adam
2. Loss = categorical_crossentropy
3. Metrics = accuracy

Using EarlyStopping

Early stopping is a method that allows you to specify an arbitrary large number of training epochs and stop training once the model performance stops improving on a hold out validation dataset.

Reducing Learning Rate on Plateau

ReduceLROnPlateau will adjust the learning rate when a plateau in model performance is detected, e.g. no change for a given number of training epochs. This callback is designed to reduce the learning rate after the model stops improving with the hope of fine-tuning model weights.

Dataset

link : https://drive.google.com/file/d/1ZSwg45cvA5KaEXmWk-OUxIeoMKY-8hs4/view?usp=sharing