Deepfake detection challenge, integration to protect content integrity unveiled

A new deepfake detection competition has been announced with the intention of advancing “next-generation deepfake detection and localization systems” development. The challenge focusses on the detection of “small fake segments” within genuine videos

The 1M-Deepfakes Detection Challenge 2024 is led by a team of seven researchers from five universities in Australia, the UK and UAE, plus Qualcomm. The researchers note in a paper explaining the challenge that most work in the field to date focusses on facial manipulation. While this is understandable as a defense against attacks on biometric and KYC systems, it leaves open a range of potential harms from misleading content.

Challenge participants will use the AV-Deepfake1M dataset, which is made up of 1 million manipulated videos of more than 2,000 subjects. The audio-visual dataset was built using a large language model (LLM)

The researchers provide baseline benchmarks, evaluation scripts and accompanying code via GitHub.

A pair of winners in Ant Group and Conference on the Bund’s Global Multimedia Deepfake Detection Challenge 2024 were announced last week.

Wolftech selects Reality Defender deepfake detection

Also working to defend against manipulated content are Reality Defender and Wolftech, with the latter integrating the former’s software to help its clients identify deepfakes.

Wolftech’s media and professional clients can use Reality Defender’s deepfake detection platform with Wolftech’s workflow management system to protect the integrity of their work and avoid spreading false information, according to the announcement.

“As AI-generated disinformation rises, access to detecting deepfakes will help those in this field cut through deceptive and disinformative content,” says Reality Defender Co-founder and CEO Ben Colman.

Reality Defender says in a blog post that its software can authenticate media or flag it as manipulated in real-time.

Article Topics

biometric liveness detection  |  biometrics  |  biometrics research  |  deepfake detection  |  deepfakes  |  Reality Defender