CIRCO

The 'CIRCO' dataset/benchmark is used to evaluate the performance of composed image retrieval (CIR) systems by providing a framework for testing zero-shot retrieval capabilities without relying on pre-collected triplets of query images, query texts, and target images.