AAI_2025_Capstone_Chronicles_Combined

4

The dataset is made out of a total of 190335 images, which are split into a Training (~75%), Validation (~20%) and Test (~5%) datasets: Figure2 Distribution of samples in each split

The only issue with this dataset is that it does not include any additional features other than whether an image is “Real” or “Fake”. This would be a problem when trying to analyze the results and compare the models, as it would be hard for to be able to pinpoint if there is any kind of bias or correlate specific image features with the performance of the models through different scenarios. Because of this, I broke down my exploratory data analysis into two stages: pixel value statistics and image quality assessment.

361