32 upvotes
New Benchmark Targets Deepfake Tricks in Long-Form Videos
What is AV-FABLE: Audio-Visual Manipulation Forensics in Long-Form Videos?
AV-FABLE introduces a dataset and evaluation framework for spotting short, localized audio-visual manipulations inside hour-long footage. It moves beyond the usual short talking-head deepfake tests, letting researchers train and benchmark models that can pinpoint fake moments in otherwise genuine videos.
Who it is for
Researchers and forensic analysts should care because it provides a realistic testbed for building tools that catch subtle deepfakes in movies, livestreams, or surveillance clips.