Title | ||
---|---|---|
ResFPA-GAN: Text-to-Image Synthesis with Generative Adversarial Network Based on Residual Block Feature Pyramid Attention |
Abstract | ||
---|---|---|
Text-to-image synthesis based on generative adversarial networks (GAN) is a challenging task. The developed methods have show prominent progress on visual quality of the synthesized images, but it still face challenge in the image synthesis of details. In this paper, we introduce an image synthesis algorithm based on semantic description and propose a residual block feature pyramid attention generative adversarial network, called ResFPA-GAN. This network introduces multiscale feature fusion by embedding feature pyramid structure to achieve the fine-grained image synthesis. The quality of the image synthesis can be improved via the iterative training of GAN, while the reference of attention can enhance the network's learning of the details of image texture. Through extensive experimental comparison on the CUB dataset, our method can achieve significant improvement on the variety and authenticity for the synthesised images. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/ARSO46408.2019.8948717 | 2019 IEEE International Conference on Advanced Robotics and its Social Impacts (ARSO) |
Keywords | Field | DocType |
multiscale feature fusion,feature pyramid structure,fine-grained image synthesis,image texture,text-to-image synthesis,synthesized images,image synthesis algorithm,residual block feature pyramid attention generative adversarial network,ResFPA-GAN,CUB dataset | Computer vision,Residual,Feature fusion,Embedding,Generative adversarial network,Computer science,Image texture,Image synthesis,Pyramid,Artificial intelligence,Generative grammar | Conference |
ISSN | ISBN | Citations |
2162-7568 | 978-1-7281-3177-1 | 0 |
PageRank | References | Authors |
0.34 | 2 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Jingcong Sun | 1 | 0 | 0.34 |
Yimin Zhou | 2 | 87 | 15.04 |
Bin Zhang | 3 | 213 | 41.40 |