CoRoI injects a chain of language-guided image regions into LLM hidden layers and reports improved MLLM benchmark scores at 7B-34B scale.
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
-
Visual Instruction Tuning with Chain of Region-of-Interest
CoRoI injects a chain of language-guided image regions into LLM hidden layers and reports improved MLLM benchmark scores at 7B-34B scale.