Microsoft Co

The Microsoft Co dataset is used to evaluate the effectiveness of adversarial attacks on vision-language models by assessing their cross-modal reasoning capabilities and decision-making biases induced by semantic manipulations.

Papers using Microsoft Co (3)

TRAP: Targeted Redirecting Of Agentic Preferences2025VisRet: Visualization Improves Knowledge-Intensive Text-to-Image Retrieval2025Intra-Modal Constraint Loss For Image-Text Retrieval2022