IaMSB applies a Schrödinger Bridge in two stages to estimate cross-modal consistency and localize deepfake intervals, reporting 3-10% gains in AP@0.95 especially on single-sided forgeries.
Language- guided contrastive audio-visual masked autoencoder with automatically generated audio-visual-text triplets from videos
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Inconsistency-aware Multimodal Schr\"odinger Bridge for Deepfake Localization
IaMSB applies a Schrödinger Bridge in two stages to estimate cross-modal consistency and localize deepfake intervals, reporting 3-10% gains in AP@0.95 especially on single-sided forgeries.