A self-supervised learning method and device based on motion sequential regression

Abstract

The invention discloses a self-supervised learning method and device based on motion sequential regression. The method includes: step S1, randomly intercepting video sample segments with a fixed number of frames; step S2, calculating and obtaining the average motion amount of the video sample segments; Step S3, randomly disrupting the sequence of frames in the video sample segment to obtain an out-of-order video sample segment; Step S4, performing sequential evaluation on the out-of-order video sample segment; Step S5, integrating the video obtained in step S2 The average amount of motion of the sample segment and the order evaluation result obtained in step S4 generate the final order score for the out-of-order video sample segment and use it as the regression target value, that is, label the label. The present invention makes full use of the information of the consistency of motion order in the video , by randomly generating samples and the order of frames within the samples, establishing a sequential evaluation standard for automatic sequential scoring evaluation, so as to achieve the purpose of automatic labeling.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a self-supervised learning method and device based on motion sequential regression. Background technique [0002] In recent years, technologies based on deep learning (Deep Learning) have achieved good results in the field of computer vision, such as face recognition and target classification. Representative deep learning methods include CNN (convolutional neural network), RNN (recurrent neural network), GAN (generative confrontation network), etc. The emergence of deep learning technology has greatly improved the accuracy of traditional recognition algorithms, but its dependence on the number of labeled samples has also increased significantly. In order to obtain an ideal model training effect, it is often necessary to provide a large amount of labeled data as training samples. Therefore, the demand for labeled samples is growing rapidly. [0003] However, sample lab...

AI Technical Summary

Problems solved by technology

There is still a lack of effective means of utilizing the time relationship

Images