Key point detection method, system and device and storage medium

Abstract

The invention provides a key point detection method, system and device and a storage medium, and the method comprises the steps: inputting a to-be-detected image into a trained activation graph generation model, and obtaining a gradient type activation graph of at least one stage; inputting a to-be-detected image and the gradient category activation graph into the trained key point detection model, and respectively inputting the gradient category activation graph into the corresponding non-local modules to obtain a key point prediction graph; and analyzing all connected regions in the key point prediction graph, and calculating the center of the connected region with the maximum area to serve as the position of the corresponding key point. The method is advantaged in that rapid and accurate key point detection is realized on the basis of the spatial attention mechanism, and the method can be applied to clothing key point detection and can also be applied to key point detection of othertypes of images in other fields.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a key point detection method, system, device and storage medium. Background technique [0002] Key point detection is one of the common tasks in computer vision. By identifying the image, it is judged whether there are key points in the image and the location of the key points, such as: human body key point detection, vehicle key point detection, clothing key point detection, etc. The key points of clothing are some special positions marked in the clothing, such as cuffs, necklines, trousers and so on. Detecting key points in clothing can play a role in the subsequent understanding of clothing, such as helping to predict clothing attributes and categories, and so on. [0003] There are two main types of existing key point detection methods. The first type is to directly use the image as input through the convolutional neural network, and at the same time directl...

AI Technical Summary

Problems solved by technology

However, the methods in the prior art still have some defects in the detection of key points, and there is still room for further improvement in the accuracy rate. At the same time, the speed of using the cyclic neural network is too slow and the efficiency is low.

Images