Implementaton of CNN on a real life dataset with the use of fast.ai library.
The model was deployed online and can be accessed here. It was trained on the web scraped images of 15 types of sneakers, such as Nike Air Force 1, Air Jordan 11 or Yeezy Boost 350. Pytorch's pre-trained ResNet34 was used, and the error rate of 0.1 on the validation set was achieved.
Play around with the website by selecting an image to upload and clicking 'analyze'
Next iterations of the project would involve adding more sneaker types and developing a more sophisticated way to handle the unknown ones.