We have a spin on a classical problem of computer vision, in which we have to classify cat and dog images using Keras.
See the jupyter notebook here.
In this Kaggle dataset, we also have a third class, named wild, that encompasses wild animals such as lions, leopards, tigers, foxes, wolves, and so on (other carnivorans overall). Therefore, instead of being a binary classification problem (dogs or cats), we have a multiclass classification problem (dogs, cats or wild).
Using a convnet with dropout and data augmentation, we obtained a 0.9675 accuracy score.
Using a pre-trained convnet (imagenet) with data augmentation, we obtained a 0.9850 accuracy score.
We also trained the dataset with other known classification algorithms (see the jupyter notebook here) such as logistic regression, SVM (Support Vector Machine), random forest and decision tree -- but to lesser results.