Love Dogs? Check it!
I was able to take my deep learning to the next level, with this handsome guy here. His name is Ranger, but sometimes I wonder what Rangers mix is as he is a good ole craigslist dog. As many dogs are not a pure breed but mixes and no one has time to do vet blood work to tell what they are, so I thought my next project would be fun for all ages to see what furr ball you could have. I used the resnet34 image classification architecture to train the model to get an error rate of 7%! An industry-standard in today's image classification architecture. The images were trained on bing search images of some of the top breeds listed on the American Kennel Club. Bing has an easy to gain access to API that allows you to easily gather up to 150 images per parameter in this case dog breed. I was able to gather 5000 images to train my model on in the process. You can find the app here in my first link and see if your good boy/girl can be classified.
Tips:
Full body shot
Good lighting
limited/no obstructions
Because the images are trained are perfectly taken pictures found on search engines such as bing these tips are very important otherwise you will get your dog classified as a cat! Not really, but still reasons being that color, light, and visibility of the puppy, in general, are the main source to recognize the image.
Also, the ability to launch the app will take a min as the file is made into a docker file by binder and has to be recreated each time.
App/ Github Repo
https://github.com/andrewdarmond/PuppyPalooza
Fastai:
https://www.fast.ai/
Binder:
https://mybinder.org/
Docker:
https://www.docker.com/play-with-docker
Who's a good pup!
Top comments (0)