Feature comparison identification method compatible with models of different versions

Abstract

The invention relates to a feature comparison recognition method compatible with models of different versions, which is characterized by comprising the following steps of: acquiring a version model to be processed and extracting features of the version model; inputting the features into a transformation network for transformation processing and output, and then cxarrying out recognition processing on the features and the features of a base library version model; wherein the formation method of the transformation network comprises the following steps that a shallow neural network is designed, a plurality of pairs of training samples are prepared, each training sample comprises features of a to-be-processed version model and features of a bottom library version model corresponding to the features of the to-be-processed version model, and the features of the to-be-processed version model are input by the shallow neural network; the features of the base library version model are used as supervision of output of the shallow neural network to calculate loss, a loss function adopts smothL1Loss, a training model is obtained, If the training model converges, the transformation network is formed. According to the method provided by the invention, on the premise that the features of the base library version model are not updated, the features extracted from the new version model can be compared with the features of the base library version model.

Description

technical field [0001] The invention relates to the fields of machine learning and pattern recognition, in particular to a feature comparison and recognition method compatible with different versions of models. Background technique [0002] One of the main applications of the deep learning model is to extract the features of the input object. Taking face recognition as an example, the deep learning model will first extract the features of the base image for storage, and the face image to be recognized will be characterized by the same model. Extract, and then compare the feature (query) with the base library feature (gallery) one by one (seeking feature similarity), retrieve the most similar feature and then perform subsequent processing. [0003] When the model is updated, not only the object to be recognized needs to be extracted with the new model, but all the features of the bottom library need to be updated with the new model for feature extraction. When the amount of d...

AI Technical Summary

Problems solved by technology

[0003] When the model is updated, not only the object to be recognized needs to be extracted with the new model, but all the features of the bottom library need to be updated with the new model for feature extraction. When the amount of data in the bottom library is huge, the cost of each iteration will increase. very high

Images