Visual similarity learning method based on depth unsupervised learning

Abstract

The invention provides a visual similarity learning method based on depth unsupervised learning, Its main contents include: Compact clustering and batch generation, Convolution Neural Network (CNN) training, local time pooling and multi-instance learning, The process is, The associated initial sample set is first obtained from the sample, the single cost function is then optimized to obtain a compact cluster (a compact and similar sample set with compact location distribution), Then, the CNN is trained alternately, and the generated similarities are subjected to local time pooling operation, and the clusters and batches are recalculated using the obtained similarities, and the similarities among the samples are obtained after many iterations. The present invention solves the problem that the prior visual similarity learning method requires a large amount of manual labeling data and has high calculation cost, can provide more fine similar structure, and has good performance in attitudeanalysis task and classification problem.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a visual similarity learning method based on deep unsupervised learning. Background technique [0002] In the field of computer vision, visual similarity refers to the degree of similarity between images or videos, and the learning of visual similarity plays a central role in many computer vision tasks involving different levels of abstraction, from low-level image processing to high-level objects recognition or human pose estimation. In the field of security, the use of visual similarity learning technology can quickly identify criminals by comparing surveillance videos and pictures of criminals in the database; it is also widely used in the field of robotics, and the use of visual similarity learning technology can make robots more accurate Identify targets and reduce the error rate; in the field of remote sensing, visual similarity learning technology can also be used to detect...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problem that the previous visual similarity learning method requires a large amount of manually labeled data and high computational cost, the purpose of the present invention is to provide a visual similarity learning method based on deep unsupervised learning, first obtain the associated initial samples from the samples set, and then optimize a single cost function to obtain compact clusters (sample groups with compact and similar location distributions), and select clusters with consistent similarities to form stochastic gradient descent (SGD) batches, and then alternately train CNNs, and Perform a local time pooling operation on the generated similar points, and use the obtained similar points to recalculate the clusters and batches, and obtain the similarity between samples after multiple iterations

Images