Mobile terminal portrait intelligent background replacement method based on mobile deep learning engine

Abstract

The invention provides a mobile terminal portrait intelligent background replacement method based on a mobile deep learning engine. The method at least comprises the following steps: S1, selecting a convolutional neural network model to be trained; s2, training a convolutional neural network model at a server; s3, deploying the convolutional neural network model at a mobile terminal based on a mobile deep learning engine in combination with an adaptive multistage model selection strategy; and S4, carrying out portrait intelligent background replacement by utilizing the selected optimal convolutional neural network model. According to the invention, the background replacement function can be realized on the mobile device, and the problems of low processing efficiency and low success rate ofportrait background replacement caused by network factors are solved; besides, when the convolutional neural network model mobile terminal is deployed, a self-adaptive multistage model selection strategy is combined, the purpose of effectively selecting the optimal model according to the user equipment difference is achieved, and the user experience is improved.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence computer vision, and more specifically, relates to a mobile terminal portrait intelligent background replacement method based on a mobile deep learning engine. Background technique [0002] Digital portrait segmentation is a basic and important research in the field of computer vision. It requires the model to recognize the human body and background in the image, that is, to perform binary classification on the image. Digital portrait segmentation has many applications, such as image description (ImageCaption), background replacement (BackgroundReplace), etc. [0003] Digital portrait segmentation is mainly based on two deep neural networks: Image Semantic Segmentation and Lightweight Backbone. The main purpose of the image semantic segmentation network is to distinguish the category of each pixel in the image. According to the given Semantics, the pixel color label (mask)...

AI Technical Summary

Problems solved by technology

In addition to network card performance factors, as far as network conditions are concerned, once there is high concurrency, high load or route damage, it is likely to cause network congestion, resulting in a significant extension of image processing time, which is very important for client applications that require high real-time interaction. is extremely serious

Therefore, almost all researches are dedicated to how to simplify the network so that it can be applied to the mobile terminal. The focus is mainly on the mobile terminal software, network and service terminals, but almost no research can truly fully realize the lightweight network. Deploying the portrait background replacement of the landing application on the mobile terminal

In addition, a small number of portrait background replacement methods that can be deployed on the mobile terminal directly select the designated device for image processing, without considering the performance difference of the mobile terminal device and the user's business needs, resulting in poor user experience.

Images