Adding MLLM-generated visual semantic descriptions, fused at instance and prototype levels, improves image-text retrieval across multiple baselines and domains.
Regionclip: Region-based language-image pretraining,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.MM 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Visual Semantic Description Generation with MLLMs for Image-Text Matching
Adding MLLM-generated visual semantic descriptions, fused at instance and prototype levels, improves image-text retrieval across multiple baselines and domains.