Our training data is comprised of 10,000+ hours of cat memes sourced from the darkest corners of the internet.
These memes are categorized into 12 different sub-genres, including:
We use a custom-built image processing pipeline to extract and normalize features from each meme, including:
We train a state-of-the-art CNN on the preprocessed data, using a custom loss function designed to maximize the probability of detecting cat memes.
Our model is trained on a mix of 80% real cat images and 20% non-cat images, carefully selected to ensure the most challenging test cases.
We test our model on a diverse set of unseen cat and non-cat images, refining the model until it achieves 99.99% accuracy.