A training and inference framework that decomposes visual queries into nested phrases and grounds them progressively outperforms baseline LVLMs on compositional grounding and reasoning benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Progressive Multi-granular Alignments for Grounded Reasoning in Large Vision-Language Models
A training and inference framework that decomposes visual queries into nested phrases and grounds them progressively outperforms baseline LVLMs on compositional grounding and reasoning benchmarks.