Dog-Cat Classifier

By Arda Mavi

Dog and cat image classifier with deep learning.

Example:


Dog: 0.92035621 Cat: 0.04618423	Cat: 0.90135497 Dog: 0.09642436

Layer outputs of test photographs:


Layer: 1 Kernel: 4	Layer: 2 Kernel: 16	Layer: 3 Kernel: 10

Look up `Data/Layer_Outputs` folder for other outputs.

Using Predict Command:

python3 predict.py <ImageFileName>

Model Training:

python3 train.py

Using TensorBoard:

tensorboard --logdir=Data/Checkpoints/./logs

Model Architecture:

Input Data Shape: 64x64x3

Layer 1:

Convolutional Layer 32 filter Filter shape: 3x3
Activation Function: ReLu
Max Pooling Pool shape: 2x2

Layer 2:

Convolutional Layer 32 filter Filter shape: 3x3
Activation Function: ReLu
Max Pooling Pool shape: 2x2

Layer 3:

Convolutional Layer 64 filter Filter shape: 3x3
Activation Function: ReLu
Max Pooling Pool shape: 2x2

Classification:

Flatten
Dense Size: 64
Activation Function: ReLu
Dropout Rate: 0.5
Dense Size: 2
Activation Function: Sigmoid

Optimizer: Adadelta

Loss: Binary Crossentropy

Adding new train dataset:

If you want to add new dataset to datasets, you create a directory and rename what you want to add category (like 'cat' or 'phone').

If you want to add a new training image to previously category datasets, you add a image to about category directory and if you have npy files in Data folder delete npy_train_data folder.

Note: We work on 64x64 image also if you use bigger or smaller, program will automatically return to 64x64.

Important Notes:

Used Python Version: 3.6.0
Install necessary modules with sudo pip3 install -r requirements.txt command.