The Zero-shot Composed Person Retrieval (ZS-CPR) task aims to jointly utilize visual and textual information for retrieving specific individuals. Traditional person retrieval methods, whether image-based or text-based, often struggle to effectively leverage both types of information, resulting in reduced accuracy. To address this, the Composed Person Retrieval (CPR) task is introduced, which combines image and text information for target person retrieval. However, supervised CPR methods require costly manual annotation datasets that are currently unavailable. To overcome this, the ZS-CPR task is proposed, which leverages existing domain-related data to solve the CPR problem without relying on expensive annotations.
cross-modal-retrieval
Zero-shot Composed Person Retrieval
3260 papers • 126 benchmarks • 313 datasets
(Image credit: Papersgraph)
Benchmarks
These leaderboards are used to track progress in cross-modal-retrieval
Trend
Dataset
Best Model
Actions
ITCPR dataset
ITCPR dataset
Libraries
iUse these libraries to find cross-modal-retrieval models and implementations
Datasets
Subtasks
No subtasks available.
Most implemented papers
Word for Person: Zero-shot Composed Person Retrieval
Fei Su, Delong Liu, Haiwen Li, Zhicheng Zhao, Hongying Meng•Sat Dec 31 2022
The Zero-shot Composed Person Retrieval (ZS-CPR) is introduced, which leverages existing domain-related data to resolve the CPR problem without reliance on expensive annotations, and a two-stage learning framework, Word4Per, where a lightweight Textual Inversion Network (TINet) and a text-based person retrieval model based on fine-tuned Contrastive Language-Image Pre-training (CLIP) network are learned without utilizing any CPR data.