MASS re-scores image-text pairs with pointwise mutual information, estimated by comparing caption likelihood on the real image versus a black image, and reduces language bias on color, counting, gender, and compositionality benchmarks.
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
-
MASS: Overcoming Language Bias in Image-Text Matching
MASS re-scores image-text pairs with pointwise mutual information, estimated by comparing caption likelihood on the real image versus a black image, and reduces language bias on color, counting, gender, and compositionality benchmarks.