Image retrieval using unsupervised prompt learning and regional attention

Identifying the target object can produce more accurate and discriminating feature representations. It can significantly improve large-scale instance-level image retrieval performance. However, it is usually difficult to obtain annotation information of all target objects in a dataset by manual annotation, which makes it challenging to automatically identify the target object. To solve this problem, we propose a novel method to recognize the target object based on unsupervised prompt learning and regional attention (PLRA) rather than manual annotation. It includes three highlights: (1) We propose an unsupervised prompt learning method to identify the target object, which can reconstruct the deep features by mining the prompt information and then design prompt factors to identify the target object based on the reconstructed features. (2) We propose a new regional attention method to extract distinguishable features of the target object. This method captures important feature regions through four dimensions: global, local, spatial and channel. It can improve the diversity and discriminability of the representation. (3) We propose a general hybrid PCA-whitening (HPW) method based on multi-parameter learning and feature fusion, to trade-off feature dimensionality and retrieval performance. This method can significantly improve the performance and reduce the vector dimensionality in a plug-and-play manner. We conducted comprehensive experiments on five benchmark datasets, and the results show that the proposed method significantly outperforms existing state-of-the-art methods based on unsupervised feature aggregation.