Training method of image generation model and image processing method and device

Abstract

The invention relates to a training method of an image generation model and an image processing method and device, and relates to the technical field of image processing. The method comprises the following steps: grouping feature layers of a feature map according to the channel number of the feature map to obtain a target feature map containing a target number of feature layer groups, setting learnable parameters of a preset normalization conversion function according to a random vector of an input model, the channel number of the feature map and the target number to obtain a target normalization conversion function, and performing normalization on the target normalization conversion function. normalizing the feature map according to the target normalization conversion function, thereby completing training of the image generation model. In the normalization operation process, both the input of the image generation model and the characteristics of the normalization operation object are considered, so that parameters in the normalization operation process are correlated, differences do not disappear due to the normalization process, the training effect of the model for image generation is improved, the quality of the generated image is improved. And the training time of the model is shortened.

Description

technical field [0001] The present disclosure relates to the technical field of image processing, and in particular to a training method for an image generation model, an image processing method and a device. Background technique [0002] With the development of technology, the application scope of deep learning is becoming wider and wider. Deep learning is to learn the internal laws and representation levels of sample data, so that the network model can have the ability to analyze and learn like a human being, and can recognize data such as text, images, and sounds. [0003] In the training process of the deep learning network model, the feature map processed by the network layer in the deep learning network model is generally normalized to obtain normalized features, so that the average value of the feature map data becomes 0, standard A distribution with a difference of 1 or a distribution in the range of 0 to 1, which can shorten the model convergence time and improve th...

AI Technical Summary

Problems solved by technology

[0003] In the training process of the deep learning network model, the feature map processed by the network layer in the deep learning network model is generally normalized to obtain normalized features, so that the average value of the feature map data becomes 0, standard A distribution with a difference of 1 or a distribution in the range of 0 to 1, which can shorten the model convergence time and improve the model training effect. Among them, the current training method uses the standardized score method, which is based on the mean and standard deviation of the original data. Standardize so that the processed data conform to the standard normal distribution, that is, the mean is 0 and the standard deviation is 1, that is, the formula is Perform normalization operation, where x is the original data, E(x) is the mean value of x, Var(x) is the variance of x, y(x) is the result of normalization processing, γ and β are learnable parameters, Among them, γ and β are independent values ​​that are independently optimized with the loss backpropagation process in the model training process and are not related to the model input and the model. This will cause the difference to disappear due to the normalization process and reduce the image quality.

Images