基于文本描述的图像生成研究(Python)

基于文本描述的图像生成研究(Python)(任务书,开题报告,论文15000字,参考代码)
摘要
生成对抗网络是图像生成方面的重点关注模型。本文主要论述了如何将文字中的描述信息正确对应到生成的图像中。系统的实现原理分为两个模块，分别是文本特征的提取和图像的生成。文本特征通过字符级卷积循环网络进行提取，以作为生成对抗网络的输入条件。生成对抗网络通过训练网络中的生成模型和判别模型，生成符合文本描述的逼真图像。为了改进图像生成的质量，研究人员探究生成对抗网络的改进方式，目前已经研究出了多种衍生模型。本文分析学习了几种对抗网络的衍生模型，但是由于条件限制，并未全部做测试。
本次课题仿真是基于Python的模拟实现，在已有图像数据集和对应标签集上进行测试，实验得到了手写数据和花卉的生成图像。但是由于训练的模型还不够稳定，图像质量并不是很好。

关键词：生成式对抗网络；文本描述；数据集；衍生模型
 
Abstract
Generative adversarial network is the focus of image generation. This paper mainly discusses how to correctly correspond the description information in the text to the generated image. The implementation principle of the system is divided into two modules, namely, the extraction of text features and the generation of images. Text features are extracted by character-level convolution recurrentneural network as the input condition for generative adversarial networks. Through training the generating model and discriminant model in the network, generative adversarial network generates realistic images that conform to the text description. In order to improve the quality of image generation, researchers have explored the improvement of generating countermeasures network, and have developed a variety of derivative models. In this paper, several derivative models of antagonistic networks are analyzed and studied, but due to the limitations of conditions, not all of them have been tested.
The simulation of this subject is based on the simulation implementation of Python, which is tested on the existing image data set and the corresponding label set. The handwritten data and flower images are obtained by the experiment.However, because the training model is not stable enough, the image quality is not very good.

Keywords: generative adversarial network; text description; dataset; derivative models
 

目录
第1章绪论    1
1.1 研究目的及意义    1
1.2 研究现状    1
1.3 预期目标和论文构成    3
第2章文本特征提取原理    4
2.1 文本表示的模型    4
2.2 文本特征提取方案    5
2.3 特征提取方法选取    9
第3章系统的设计与实现    10
3.1 生成对抗网络GAN    10
3.2 GAN-CLS和GAN-INT算法改进    12
3.3 衍生模型    13
3.3.1 CGAN    13
3.3.2 DCGAN    14
3.3.3 WGAN    15
3.3.4 StackGAN    16
3.3.5 PGGAN    18
3.4 GAN相关模型选择测试    19
3.5 测试数据集    19
3.6 测试结果及分析    20
第4章总结与展望    25
参考文献    26
致谢    28