In the Cats v Dogs project, the neural network was trained to tell apart just 2 classes : Cats and Dogs. Here we try something more ambitious with a Kaggle dataset containing 200 species of birds.
Birds can be tiny like a Hummingbird, large like an Eagle or truly large like a California Condor. Even with this variety, the model predicts with an accuracy of about 75%. That's not bulletproof but compared to a random guess that has a 0.5% chance of getting the right speices out of 200, 75% isn't too shabby! Having said that, remember that there are nearly 10,000 bird species on planet Earth. This model has been trained on a tiny fraction i.e. 2% of all bird species and will end up classifying any bird it sees as one of these 200 - the closest match. Still, for dozens of generally well-known species such as Eagles, Swans, Peacocks, etc. it does quite well.