Reproduce “AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhixi Cai, Shreya Ghosh, Aman Pankaj Adatia, Munawar Hayat, Abhinav Dhall, Tom Gedeon, Kalin Stefanov
AV-Deepfake1M presents a large audio-visual benchmark for locating short manipulations within otherwise genuine videos. It contains 1,146,760 videos from 2,068 subjects, with word-level replacements, deletions, and insertions across audio and video. The authors benchmark existing methods for temporal localization. UMMAFormer reports [email protected] 98.83 on LAV-DF and 51.64 on AV-Deepfake1M, showing lower reported performance on this synthetic VoxCeleb2-derived dataset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.