A unified reference-free faithfulness metric for video-to-text and text-to-video that uses fact decomposition, semantic dependency graphs, and VideoQA models.
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
-
FIFA: Unified Faithfulness Evaluation Framework for Text-to-Video and Video-to-Text Generation
A unified reference-free faithfulness metric for video-to-text and text-to-video that uses fact decomposition, semantic dependency graphs, and VideoQA models.