Cats VS Dog Classifier with Tkinter GUI

Pre-requisites:

• Python3
• pandas
• numpy
• NVIDIA® GPU card with CUDA® Compute Capability 3.5
• NVIDIA® GPU drivers —CUDA 10.0 requires 410.x or higher.
• CUDA® Toolkit —TensorFlow supports CUDA 10.0 (TensorFlow >= 1.13.0)
• CUPTI ships with the CUDA Toolkit.
• cuDNN SDK (>= 7.4.1)
• Tensorflow

You can visit Tensorflow website, for installation guide for above requisites.

About Dataset

• The train folder contains images of cat and dogs.The train folder contains 25,000 images of dogs and cats.There are no duplicate images.
• The test folder contains 12500 images of cats and dogs.
• Each image has been preprocessed to 200x300 pixels.
• test Dataset in unlabelled.
  URL for dataset : https://www.kaggle.com/c/dogs-vs-cats/data

About Model

• I have used Convolutional Neural Network to classify images.
• Here is the architecture of the model.
• 
  
• The model has loss: 0.0679, acc: 0.9750, validation loss: 0.3017 - validation acc: 0.9016
• After Training I saved the model so that I could use it without training the model again and again
• Refer to Cat Vs Dog.ipynb to checkout all the details and code.

Cats VS Dogs GUI

• I developed the GUI using Python inbuilt module Tkinter.

GUI of the Application

• There are 4 buttons on the window

  • Browse files - This button can be used if you want to select multiple images.

  • Browse directory - This button can be used if you want to test accuracy of the algorithm on multiple images. You must select a directory using this button.

  • Predict - This Button is used after selecting multiple image files(This is used with Browse Files, See same color of the button represents they are used together).

  • Predict Accuracy - This button is used after selecting the directory of images(This is used after Browse directory).The directory names should be cats and dogs as shown in screenshot.

Browse files Button Action

Predict Button Action

Browse Directory Button Action

Predict Accuracy Button Action

Preprocess Images

• To show real-time results to used I implemented preprocessing of images so that the algorithm is able to work on all image sizes.
• Every image is reshaped to 300x400 resizing.

Executing the model

• First make sure that all the pre-requisites are installed in your computer.
• Then execute the 'CatsVsDogs GUI.py' using :-
  python 'CatsVsDogs GUI.py'
  OR
  python3'CatsVsDogs GUI.py'