Hand-painted 3D building automatic coloring network structure and method based on ArcGAN network

Abstract

The invention discloses a hand-painted 3D building automatic coloring network structure and method based on an ArcGAN network, and the network structure comprises the ArcGAN network; the ArcGAN network is composed of a first generator and a first double discriminator, and the first double discriminator comprises a first global discriminator and a first local discriminator; the network structure further comprises an Attention module and an LSTM module. The Attention module is composed of a second generator and a second discriminator; the second generator generates a picture considered to be real by the second discriminator according to the input picture, and the second generator is composed of a second encoder, a second converter and a second decoder; the second discriminator is used for discriminating whether the input picture is a composite picture from the second generator or a real picture from the training set, and is composed of continuous down-sampling convolution layers; the LSTM module is composed of a third generator and a third double discriminator; and the third generator adopts an encoder-LSTM-decoder structure, and is divided into three parts, namely a third encoder, arecurrent neural network module and a third decoder.

Description

technical field [0001] The present invention mainly relates to the field of deep learning, in particular to an ArcGAN network-based automatic coloring network structure (AL-GAN) for 3D buildings and a method thereof. Background technique [0002] Since the 1990s, research on style transfer has begun to appear in the field of deep learning. The so-called image style transfer refers to the technique of using algorithms to learn the style of famous paintings, and then applying this style to another picture. A large number of researchers have begun to study how to combine the style on one image with the content on another image to generate a new painting that has never been seen before. Gatys [2] proposed the neural style transfer algorithm (NST) in 2015, which uses a convolutional neural network (CNN) to separate and then combine the content and style of any image to generate artistic images with high perceptual quality. It has aroused widespread attention in the academic ci...

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to overcome the deficiencies in the prior art, and aims to solve the problem of style migration and continuous coloring from 3D line architectural drawings to different styles of colored architectural drawings

In order to solve the above problems, the present invention provides a hand-painted 3D building automatic coloring network structure (AL-GAN) and method based on the ArcGAN network, by building and training a deep neural network structure suitable for the data set to automatically color the line building to achieve style Migration, which solves the problem of uneven coloring and poor continuity of the image, the effect is better and it is easy to use

Images