A new benchmark, SORCE-1K, tests text-to-image retrieval of small non-salient objects, and a multi-embedding MLLM method with regional prompts outperforms single-embedding baselines on it.
X-detr: A versatile architecture for instance-wise vision-language tasks
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.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
SORCE: Small Object Retrieval in Complex Environments
A new benchmark, SORCE-1K, tests text-to-image retrieval of small non-salient objects, and a multi-embedding MLLM method with regional prompts outperforms single-embedding baselines on it.