samspre / Doggone-It

Project for AI in Fact and Fiction Summer 2019

Doggone It

Detecting Dog Breeds through Computer Vision

To format the training images, run create_training-data.py*. This pulls images from a directory that contains fodlers labeled with the respective dog breed name. Inside these folders are images of the repective dog breed. Our model is currently trained on the Standford Dog Dataset. Using Tensorflow and Keras, this program resizes the images to be 90x90 and creates an RGB shape array. The image arrays and their corresponding labels are saved to a file using pickle.

To create and train the model, run create_and_train_model.py*. The model uses the image arrays and labels to train. The images are passed through 4 Dense layers, flattened, and then passed through an output layer. The model is saved to a file using keras' .save() function.

To use the trained model, run 3-predict.py. This program follows the keras test/prediction techniques. Our GUI makes it easy for users with no coding experience to test our model. By using a location feature, we can also search for lost dogs within 500 miles of the user by utilizing online databases.

At the current moment, our model was trained with the Stanford Dog Data Set. This set contains 120 different dog breeds and 150 pictures of each breed, for a total of 18,000 training photos. We trained our model with these 18,000 photos over 20 epochs. Training took about 2 minutes per epoch, and reached an estimated 96% accuracy. Of course, with cleaner images and more iamges per breed, we could increase the accuracy of the model.

Overall, our group has learned how difficult it is to make a reliable AI. One needs a lot of quality data and a well-structured model, as well as a lot of time to create and train the model. It certainly was more difficult than we imagined.

Table of Contents

1. Members
2. Getting Started
3. Setup
4. Using the App

Members

• Virginia Barnes
• Patrick Gilbert
• Kristofer Kwan
• Eric Miu
• Jinli Park
• Samantha Sprecace

Getting Started

These instructions utilize Bash on Ubuntu for Windows with Git functionality installed

Installation

Install Python 3.6

Check to see if Python 3.6 is already on your system by typing the following into your preferred shell

python3.6 --version

If you have the latest version (as of this writing), you should see

Python 3.6.9

If Python 3.6 is not installed, update packages if needed and install

sudo apt-get update

sudo apt-get install python3.6

Upgrade packages if needed with

sudo apt-get upgrade

Install Tensorflow and Keras

Using pip3 from Python3.6

pip3 install tensorflow --user

pip3 install keras --user

Setup

1. Clone the Project

Perform the following commands to access the repository

user@computer:~$ git clone https://github.com/samspre/Doggone-It
user@computer:~$ cd Doggone-It
user@computer:~/Doggone-It$ git pull

2. Set up the virtual environment

# Windows
user@computer:~$ .venv/Scripts/activate
# Mac/OS/Linux/WSL:
user@computer:~$ source venv/Scripts/activate

Install the requiresments file, which places all the python dependencies into your virtual environment

(venv) user@computer:~/Doggone-It$ pip3.6 install -r requirements.txt --user

3. Run the program

(venv) user@computer:~/Doggone-It$ python3.6 main.py
* Serving Flask app "main" (lazy loading)
* Environment: production
  WARNING: Do not use the development server in a production environment.
  Use a production WSGI server instead.
* Debug mode: on
* Running on http://127.0.0.1:5000/ (Press CTRL+C to quit)
* Restarting with stat
* Debugger is active!
* Debugger PIN: XXX-XXX-XXX

App is now running on the given URL (may be different) To quit, type

deactivate

into the terminal

Using the App

Identification

Import your photos and go!

Lost and Found