AAI_2025_Capstone_Chronicles_Combined

3 and some recommendations on how to effectively deploy them in an actual product and some advice for future improvement. Dataset Summary The “Deepfake and Real Images” dataset is a cleaned-up version of the OpenForensics dataset by Le et al. (2021). It contains a collection of 190,000 face images, half real and half fake, with varying quality, in different scenarios. The fake images have been generated through various neural models, mainly Generative Adversarial Networks (GANs) (Goodfellow et al., 2014b). The dataset comes already divided into three splits, Training, Test and Evaluation, with Test images having been augmented to introduce different and more challenging scenarios. The quality of the fakes also varies from sample to sample, ranging from images that are easy to identify as fake to ones that are hard for humans to spot. The following image shows a random sample of images: Figure1 Sample of Fake and Generated images from the dataset

360