Image recognition model update method and related device

Abstract

The present invention relates to the field of image recognition, and provides a method for updating a recognition model and a related device. The method includes: acquiring first training data and a first recognition model trained by the first training data, and the first recognition model includes feature extraction. Network and the first classification network, the feature extraction network includes the first parameter; obtain the second training data, wherein, the second training data and the first training data contain the same type of target; keep the first parameter and the first classification of the feature extraction network The parameters of the network remain unchanged, a second parameter is added to the feature extraction network, and the second parameter is updated to obtain a second recognition model. In the present invention, while keeping the parameters of the original recognition model unchanged, the parameters of the original recognition model are added, and only the newly added parameters are trained and updated, while the recognition effect of the original classification task is kept unchanged, The training time and training resources consumed by updating the recognition model are reduced.

Description

technical field [0001] The present invention relates to the field of image recognition, in particular to a method for updating a recognition model and a related device. Background technique [0002] At present, the usual practice of image recognition tasks based on deep learning is that when recognizing and classifying images from new fields different from the current existing recognition models, it is first necessary to collect a large amount of task data in new fields, and pass A large number of manual annotations, and then model training. In order to ensure that the retrained model has a better recognition effect on the original classification task and the classification task in the new field, the method usually used in the existing technology is to mix the original task data with the task data in the new field. The entire model is retrained, which consumes a lot of training time and resources. How to reduce the training time and the consumption of training resources wi...

AI Technical Summary

Problems solved by technology

In order to ensure that the retrained model has a better recognition effect on the original classification task and the classification task in the new field, the method usually used in the existing technology is to mix the original task data with the task data in the new field. The entire model is retrained, which consumes a lot of training time and training resources

Images