CIRR

The CIRR dataset contains over 36,000 pairs of crowd-sourced, open-domain images and human-generated modifying text, and it is used to evaluate composed image retrieval in rich visual and language contexts.

Papers using CIRR (11)

ConeSep: Cone-based Robust Noise-Unlearning Compositional Network for Composed Image Retrieval2026 · 1 citesDAFM: Dynamic Adaptive Fusion for Multi-Model Collaboration in Composed Image Retrieval2025Image Retrieval on Real-life Images with Pre-trained Vision-and-Language Models2021 · 8 citesPic2Word: Mapping Pictures to Words for Zero-shot Composed Image Retrieval2023 · 2 citesKnowledge-Enhanced Dual-stream Zero-shot Composed Image Retrieval2024 · 2 citesiSEARLE: Improving Textual Inversion for Zero-Shot Composed Image Retrieval2024 · 2 citesComposed Image Retrieval using Contrastive Learning and Task-oriented CLIP-based Features2023 · 1 citesLanguage-only Efficient Training of Zero-shot Composed Image Retrieval2023 · 1 citesImproving Composed Image Retrieval via Contrastive Learning with Scaling Positives and Negatives2024 · 1 citesImagine and Seek: Improving Composed Image Retrieval with an Imagined Proxy2024 · 1 citesTraining-free Zero-shot Composed Image Retrieval via Weighted Modality Fusion and Similarity2024