CIRCO Dataset

Composed Image Retrieval on Common Objects in context

This is the evaluation server for the CIRCO dataset: the first dataset for CIR with multiple ground truths. For more details on CIRCO, please refer to our ICCV2023 paper Zero-Shot Composed Image Retrieval with Textual Inversion and the extended version iSEARLE: Improving Textual Inversion for Zero-Shot Composed Image Retrieval.

We did not release the annotations for the test set. To evaluate your model, use the buttons below.

Evaluation
How-to

This site is maintained alongside SEARLE and CIRCO. If you have any questions, please open an issue or send us an email.