CIRR

The 'CIRR' dataset/benchmark contains triplets of reference images, textual descriptions of desired modifications, and target images, and is used to evaluate the effectiveness of models in the task of Composed Image Retrieval (CIR).

Papers using CIRR (6)

MCoT-MVS: Multi-level Vision Selection by Multi-modal Chain-of-Thought Reasoning for Composed Image Retrieval2026ReCALL: Recalibrating Capability Degradation for MLLM-based Composed Image Retrieval2026WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval2026CIR-CoT: Towards Interpretable Composed Image Retrieval via End-to-End Chain-of-Thought Reasoning2025TMCIR: Token Merge Benefits Composed Image Retrieval2025CoLLM: A Large Language Model for Composed Image Retrieval2025