Code for AI or Not Competition.

Model training and evaluation code for Hugging Face AI or Not Competition. Primary objective is to achieve minimal cross entropy loss for binary image classiciation task of determining whether an image is AI generated or not. All model finetuning is performed on a single GPU through Colab. The competition dataset is used to finetune VIT-base, VIT-large, and Facebook's VIT mae (large and base) transformers. Best log loss achieved: $0.0739$ with ~97% accuracy, placing 35/98 entries.