Bark, Bark!

The project

The project is centered around image classification and the ability to use deep learning to acheive it. Today, more things are going virtual and are on phones, computers, tablets, etc very easily but some things are just better real. An extremely popular subject for entertainment of many people in real life is animals, but not ever animal can be ‘Man’s Best Friend’. Obviously, dogs love the ladies too! Choosing a dog is very touchy subject for some as breeding can make dogs cost in the thousands, but with the help of your local ASPCA, adopting your new Good Boy/Girl is just a short drive away. So, how can I tell what kind of dog I have after I adopt one without going to the vet for blood work? Easy! Use my Puppy Palooza App. I have created an app with the help of Deep Learning to achieve 90%+ accuracy across the top ~30 breeds found today using the American Kenel Club. Using Fastai, Google Collab, Bing Search API, & Endless Furr Balls. You can see what kind of breed your current dog is, a potential addition to the family, or simple curiosity. Remember it only kills cats!

Binder App

In order to make the app functional online, it is best to run the application as a container file. Binder is a new software that allows Jupyter Notebooks to operate as container files and be utilized more interactively as a web application with the power of python packages as Ipywidgets.

App in Action

Here is to the app in action via link posted on my Github.

Insights

1. The model is state of the art when classifying dog photos when they are in the same mold as found on typical seach engines, but it is not at high quality when using base photos that are from someones phone

2. Mix dogs will have lower then 50% most likely as the mixture will create the model to split the percentages based on the dog multiple breeds

3. Puppies are super cute

Conclusion

The App is fun and easy to use, but most importantly, who doesn’t like looking at puppy images all day!