Model training method and device and storage medium

Abstract

The embodiment of the invention discloses a model training method and device and a storage medium. The method includes: obtaining a multi-label image training set; selecting a target training image currently used for model training from the multi-label image training set; carrying out label prediction on the target training image by adopting a deep neural network model to obtain a plurality of prediction labels of the target training image; obtaining a cross entropy loss function corresponding to the sample label, the positive label loss in the cross entropy loss function being provided with aweight, and the weight being greater than 1; and converging the prediction label and the sample label of the target training image according to the cross entropy loss function to obtain a trained deep neural network model. According to the scheme, the model accuracy and visual expressive force can be improved.

Description

technical field [0001] The present invention relates to the field of communication technology, in particular to a model training method, device and storage medium. Background technique [0002] With the development of deep learning models and training methods, great progress has been made in the field of computer vision, and the research direction has gradually shifted from low-level image processing and image recognition to higher-level visual understanding. Complex vision tasks require the utilization of deep neural network models with the potential for better visual representation. [0003] At present, the deep neural network model trained on the large-scale multi-label image training set has better visual expression ability, and the quality of the large-scale multi-label image data set determines the visual expression and accuracy of the deep neural network model. . The currently public large-scale multi-label image dataset ML-Images can include 11,166 labels and 18,01...

AI Technical Summary

Problems solved by technology

[0004] However, the current large-scale multi-label image data set has a large category imbalance problem, for example, the problem of imbalance between positive and negative labels within the category, for example, for a certain training image in the data set, generally its negative label (ie categories that do not exist in this image) are much larger than the number of positive labels (categories that do not exist in this image)

Due to the problem of category imbalance in large-scale multi-label image datasets, the accuracy and visual expressiveness of deep network models trained on large-scale multi-label image datasets are low.

Images