Image recognition model training method, electronic device and storage medium

Abstract

The embodiment of the present application relates to the field of visual search technology, and discloses a training method for an image recognition model, an electronic device, and a storage medium. The method includes: obtaining a training sample of the first model; wherein, the training sample is marked to represent the training sample The label of the feature category of the feature; based on the network structure of the first model, construct the second model; according to the training sample and the third model, obtain the category center vector of each feature category corresponding to the third model; wherein, the first model and the third The model is a model with the same function; according to the category center vector of each feature category, the classification layer weight of the second model is determined; the second model is iteratively trained according to the training samples and labels, and the second model is updated except for the classification layer weight Parameters, the feature set extracted by the trained model can be directly compared with the feature library extracted by the old model, saving time and effort, reducing costs, and greatly improving the convenience of industrial deployment of the model.

Description

Image recognition model training method, electronic device and storage medium technical field Embodiments of the present application relate to the technical field of visual matching and search, in particular to the training of a kind of image recognition model Methods, electronic devices, and storage media. Background technique [0002] With the increasing maturity of visual matching and search technology, recognition models based on visual matching and search technology are widely used. It is widely used in many fields, such as image retrieval, pedestrian re-identification, vehicle re-identification, face recognition, etc., which are based on visual matching. The recognition model of matching and search technology maps the image to a feature embedding space through a deep neural network, where the feature space is empty. From time to time, the features of the same category are close to each other and clustered together. Generally speaking, for large-scale image data i...

AI Technical Summary

Problems solved by technology

[0003] In practical application scenarios, in order to provide users with a better experience, the recognition model based on visual matching and search technology needs to be iteratively updated. However, after the recognition model is updated, the features in the probe are extracted with the new model. However, the features in the gallery are all extracted with the old model. In order to ensure the consistency of the features between the probe and the gallery, technicians need to use the new model to re-extract features from the original image data corresponding to the gallery. This process takes a very long time. , The cost is very high, and in some scenarios with high security requirements, the original image data corresponding to the gallery is automatically deleted after the gallery is generated, and the features cannot be re-extracted, so the consistency of the features between the probe and the gallery cannot be guaranteed. cannot be compared directly

Images