Retrieval method and device based on segmentation difficulty sample generation

Abstract

The embodiment of the invention provides a retrieval method and device based on segmentation difficulty sample generation, and the method comprises the steps of increasing the difficulty degree of each original ternary image group in a sample set of an original ternary image group through employing all samples in the sample set of the original ternary image group; and in a first stage of the THSG,increasing the difficulty degree of the positive sample pair, so while the difficult positive sample pair is obtained, it is guaranteed that the label of the difficult positive sample pair is consistent with the label of the original positive sample pair; increasing the difficulty degree of the original negative sample in the second stage, obtaining the final difficult negative sample and the final difficult positive sample pair, and improving the effective usability of the sample set. Furthermore, a final difficult ternary sample group is used, and effective difficult samples can be supplemented for fewer training sets, so that the model can be better trained. Meanwhile, a more robust and robust feature extraction retrieval model is obtained through training by using difficult sample pairs.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a retrieval method and device based on generation of segmented difficult samples. Background technique [0002] Deep Metric Learning (DML) methods aim to learn powerful metrics to accurately and robustly measure the similarity between data. At present, the development of DML enables it to be widely used in various fields, such as image retrieval, person re-identification, clustering and other multimedia tasks. [0003] The above-mentioned image retrieval is taken as an example for description. At present, there are many image retrieval methods based on DML. There is mainly a method of building a model based on metric learning. In metric learning, multiple sets of triplet image group samples are used as input for building a model. Each set of triplet image group samples is composed of a pair of positive samples with the same label and a negative sample with a different ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the embodiments of the present invention is to provide a retrieval method and device based on segmented difficult sample generation to solve the problem of the limited number of triplet image group samples that can be constructed in some small-scale data sets in the prior art, making the model Technical problems that cannot be effectively trained, thereby reducing the effectiveness of retrieval

Images