DeepFake Detection Challenge

In two recently published blog posts, my colleague Michael Lomnitz and I discuss IQTLabs’s work in FaceBook’s DeepFake Detection Challenge (DFDC).

In the first post, we outline the deepfake problem, the crowd-sourcing approach of the challenge itself, and set the stage for results. The second post goes in depth on the steps for using the dataset, the models trained for deepfake detection, and finally a comparative analysis of the results.

Ultimately, the highest performing submissions were those that relied heavily on the choice of data augmentation methods rather than implementing the latest and greatest model architecture. And despite the small team, the IQTLabs solution landed in the top 28% of challenge submissions. Hope you enjoy!