Dog_Breed_Classifier

Identify dog breed using convolutional neural network(CNN)

Udacity Data Scientist Nanodegree Capstone Project

Check the descriptive Medium blog post here.

Project Motivation

This project builds up three CNNs and an algorithm to identify the breed of dogs in images and differentiate dogs with humans as well as other objects/species.

Installation

1. Install Anaconda if you have not installed it. Otherwise, skip this step.

2. Update Anaconda by typing conda update -all in Anaconda Prompt.

3. Clone this repository to your local machine using:

$ git clone https://github.com/YueminLi/Dog_Breed_Classifier.git

4. Download the following pre-trained model and save it to bottleneck_features in the Dog_Breed_Classifier repository:

• ResNet-50 bottleneck features

5. Create a new folder named saved_models in the Dog_Breed_Classifier repository.

6. Run this Jupyter notebook in your local Anaconda Jupyter Notebook environment:

dog_app.ipynb

File Descriptions

• haarcascades: folder contains pre-trained face detectors

  • haarcascade_frontface_alt.xml: the pre-trained face detector used in the dog_app.ipynb notebook

• images: folder contains all example images shown in the dog_app.ipynb notebook

• test: folder contains all test images used to test the algorithm in step 7

• dog_app.ipynb: Jupyter notebook that contains full codes to build up CNNs and algorithm

• extract_bottleneck_features.py: Python function to extract bottleneck features

• bottleneck_features: folder contains pre-trained models

  • DogVGG16Data.npz: pre-trained VGG16 model (this has already been downloaded)
  • DogResnet50Data.npz: pre-trained ResNet-50 model (this will need to be downloaded from the link provided above)

• saved_models: folder contains trained models from the dog_app.ipynb notebook (all are saved while running the dog_app.ipynb)

  • weights.best.from_scratch.hdf5: stores the trained model with the best validation loss from self-created CNN
  • weights.best.VGG16.hdf5: stores the trained model with the best validation loss from VGG16 model
  • weights.bes.Resnet50.hdf5: stores the trained model with the best validation loss from ResNet-50 model

• LICENSE

• README.md

Author

Yuemin Li

Github: https://github.com/YueminLi

LinkedIn: https://www.linkedin.com/in/yuemin-li-89166333/

License

Usage is provided under the MIT License. See LICENSE for the full details.