Picture retrieval method and device and storage medium

Abstract

The invention provides a picture retrieval method and device and a storage medium. The method comprises the steps that a to-be-tested picture with attribute label information is acquired, wherein theattribute label information comprises attribute types and attribute values corresponding to the attribute types; and inputting the picture to be detected into the trained feature extraction network toobtain a retrieval feature vector. The trained feature extraction network extracts a first retrieval feature map from the to-be-detected picture, predicts attribute confidence corresponding to each piece of attribute tag information according to the first retrieval feature map, and obtains a second retrieval feature map based on the attribute confidence and the first retrieval feature map; A retrieval feature vector is obtained based on the second retrieval feature map. According to the retrieval feature vector, a target picture matched with the to-be-retrieved picture is obtained from a to-be-retrieved data set. According to the invention, the accuracy of a picture retrieval result is improved.

Description

technical field [0001] The present invention relates to the field of computer application technology, in particular to a picture retrieval method, device and storage medium. Background technique [0002] Image retrieval is a method of extracting feature vectors from pictures or video screenshots to be tested, and then searching for query targets based on the feature vectors in a data set to be retrieved composed of a large number of pictures. The main part of the image retrieval method lies in the extraction of feature vectors. At present, the extraction method based on deep learning has become the mainstream of image retrieval feature extraction method because of its fast, efficient and strong adaptability. [0003] However, there are still some shortcomings in the existing deep learning methods: the features extracted by the current method are not enough or even difficult to focus on the attributes that the user needs to pay attention to, and the attributes that the user ...

AI Technical Summary

Problems solved by technology

[0003] However, there are still some shortcomings in the existing deep learning methods: the features extracted by the current method are not enough or even difficult to focus on the attributes that the user needs to pay attention to, and the attributes that the user does not want to pay attention to will interfere with the retrieval results

(2) Objects with non-key attributes will interfere with the extraction of information and eventually lead to retrieval errors

For example, some non-motor vehicles in the training data have drivers and some do not have drivers, which leads to overfitting of the network model and considers the driver as one of the retrieval targets, which interferes with the model training.

All of the above lead to poor accuracy of search results

Images