AI Art Detection
CNN classifier that distinguishes AI-generated images from human-made art, built during COSMOS @ UC San Diego.
The live demo could not be reached from this network.
It runs on a separate server, which some school and office networks block. The write-up below covers how it works , and the code and model files are in the repo.
Final project for COSMOS Machine Learning Cluster at UC San Diego. The model can classify if an image is AI generated or human made by identifying poor and rich contrasts within the image pixels.
The model was trained using different algorithms, like logistic regression, decision trees, and neural networks. The final model was the MLP neural network that was adjusted with CNN augmentation. Each layer of the neural network was trained on a specific subject, such as color statistics, pixel gradients, and texture variants. The final model achieved a 98% accuracy on held out testing data.
The source code, training notes, and model files are available in the GitHub repo .