Skip to content

Capstone Project 2 for Data Science Career Track: Image Classification of Dog Breeds

Notifications You must be signed in to change notification settings

thegarrickchu/Springboard-Dog_Breed_Classifier

Repository files navigation

Greetings!

This is the repository for my Capstone Project 2 for the Springboard Data Science Career Track. The project/report/deck is entitled "Who’s a good dog?

An exploration and assessment of using Convolutional Neural Networks to classify dog breeds from images".

In this repository, you will find the following folders and files:

  1. Main Repo

    1. Final Report

    2. Accompanying Slide Deck

    3. Project 2 Proposal

  2. Jupyter Notebooks:

    1. Directory Setup:Verify compressed dataset files, create train test directories and crop images according to annotation. Pickling and packaging images into single Numpy array.

    2. dog_breed_EDA: Exploring a small selection of images, validating data, sizing up the dataset.

    3. First iteration of building CNN from scratch (simple_CNN.ipynb): Initial iterations of CNN models from scratch (before Directory Set up was performed).

    4. CNN from Scratch using Pre-processed image data (simple_CNN_preprocessed_images.ipynb): Final iteration of CNN models from Scratch which became the baseline reference for pre-trained models.

    5. Using Transfer Learning (vgg_model.ipynb): Using VGG16 model architecture and pre-trained weights. Includes visualizations and predictions on test and real-world data.

    6. Deprecated:(simple_CNN_smaller_class_subset.ipynb) Tried using CNN from scratch using flow_from_directory on smaller subset of data.

    7. Deprecated: simple_CNN_v2.ipynb

=======