Training Method and Device for an Image Enhancement Model, and Storage Medium

摘要

The present invention relates to a training method, apparatus and storage medium for an image reinforcement model, wherein the training method for the image reinforcement model inputs each training input image group into an image reinforcement model, and generates a predicted image output by the image reinforcement model. obtaining; training the image enhancement model until convergence, using a loss function respectively corresponding to each training pair; The loss function includes a plurality of gray scale loss components corresponding one-to-one to a plurality of frequency sections, and each gray scale loss component includes a gray scale frequency division image of a predicted image within each frequency section, and a gray scale frequency of the corresponding target image. It is determined based on the difference value between the divided images, and different gray scale loss components correspond to different frequency sections. In the present invention, by making the loss function reflect the prediction image in the corresponding training pair, and the detailed content information and semantic information of the target image, the problem of excessive smoothing caused by inappropriate problems in the training process of the image enhancement model is effectively alleviated make it