Model training method and device, key point positioning method and device, equipment and medium

Abstract

The invention discloses a training method of a human body key point positioning model, a human body key point positioning method and device, equipment, a medium and a program product, and relates to the field of artificial intelligence, in particular to deep learning and computer vision technologies. The implementation scheme comprises: determining an initial positioning point of each key point by using a first sub-network of a positioning model; obtaining N candidate points of each key point according to the position of the initial positioning point; extracting a second feature graph of the sample image by using a second sub-network of the positioning model, and forming N groups of graph structure feature data according to the second feature graph and the candidate points; performing graph convolution on each group of graph structure feature data by using a second sub-network to obtain the offset of each key point of N groups; correcting the initial positioning point by using the offset to obtain positioning results of the N groups of key points; and respectively calculating loss values according to the marked truth value of each key point and each group of positioning results, and performing supervision training on the positioning model according to each group of loss values. According to the invention, the accuracy and robustness of key point positioning are improved.

Description

technical field [0001] This application relates to the field of artificial intelligence, in particular to deep learning and computer vision technology, and in particular to a training method for a human body key point positioning model, a human body key point positioning method, a device, a device, a medium, and a program product. Background technique [0002] At present, the main implementation scheme of human body key point positioning is usually to use deep neural network as a feature extractor, and then generate a heat map of each key point based on the feature map, and roughly locate the key point according to the maximum response position on the heat map. [0003] However, when the key points of the human body in the image are occluded, the features are not obvious, or there are other more similar interference areas, it is difficult to locate using the existing technology, or positioning errors occur. Contents of the invention [0004] The present application provide...

AI Technical Summary

Problems solved by technology

[0003] However, when the key points of the human body in the image are occluded, the features are not obvious, or there ...

Images