A two-stage audio-visual deepfake detector, HOLA, uses 1.81M pre-training samples and hierarchical cross-modal fusion modules to achieve first place and near-perfect AUC on AV-Deepfake1M++ video-level detection.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
HOLA: Enhancing Audio-visual Deepfake Detection via Hierarchical Contextual Aggregations and Efficient Pre-training
A two-stage audio-visual deepfake detector, HOLA, uses 1.81M pre-training samples and hierarchical cross-modal fusion modules to achieve first place and near-perfect AUC on AV-Deepfake1M++ video-level detection.