Dog Breeds AI

🐕 This repository contains the application that allows users to perform a breed classification of a dog image.

Example

Configuration

Dataset

The dataset that has been used is the Dog Breeds Dataset. Which contains 70 classes, split into train, validation, and test folders. If you wish to train the model on your own, you can download the dataset, and specify it's path in the application. Just change the DATA_PATH's variable value in the Classifier.ipynb file. Example

DATA_PATH = "C:/Users/Moez/Desktop/Dog_Breeds"

Installing packages

After doing so, it is recommended to create a new environment, and to run the following command to install the necessary python packages.

pip install -r requirements.txt

Usage

To use the web app directly, just run this command in a desired terminal and open the hosted app address in a browser.

python app.py

Training

Training took approximately 45 minutes to complete, with a batch size of 16. The model has been trained over 20 epochs, it has an accuracy of 93% on the validation set.

Convolutional Neural Network

• The deep learning model that has been used, is a pretrained MobileNet model. The pretrained MobileNet model has been chosen for speed and simplicity reasons.
• The final linear layer has been removed, in order to add a classifier model, which contains :
  • A GlobalAveragePooling2D layer,
  • A Dropout of 0.2,
  • A Final Dense layer of size 70 (Number of dog breeds).
• The learning rate is 0.00001,
• The optimizer in use is the Adam optimizer.