I made a dataset that modifies a deepfake classification set based on social media platform re-encoding.

When images are run through social media platforms, they are resized, re-encoded, and pushed through the platform's codec.

In assessing a deepfake detector model, it's important to ensure the model remains robust across real world platforms.