A GPT-2 mapper over dual visual encoders claims 16% training cost and better alignment, but test-time use of true class labels makes the comparison invalid for V2A.
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
REJECT 1representative citing papers
citing papers explorer
-
Efficient Video-to-Audio Generation via Multiple Foundation Models Mapper
A GPT-2 mapper over dual visual encoders claims 16% training cost and better alignment, but test-time use of true class labels makes the comparison invalid for V2A.