Synthesizing images or texts automatically becomes a useful research area in the artificial intelligence nowadays.Generative adversarial networks(GANs),proposed by Goodfellow,et al.in 2014,make this task to be done mo...Synthesizing images or texts automatically becomes a useful research area in the artificial intelligence nowadays.Generative adversarial networks(GANs),proposed by Goodfellow,et al.in 2014,make this task to be done more efficiently by using deep neural networks(DNNs).The authors consider generating corresponding images from a single-sentence input text description using a GAN.Specifically,the authors analyze the GAN-CLS algorithm,which is a kind of advanced method of GAN proposed by Reed,et al.in 2016.In this paper the authors show the theoretical problem with this algorithm and correct it by modifying the objective function of the model.Experiments are performed on the Oxford-102 dataset and the CUB dataset to support the theoretical results.Since the proposed modification can be seen as an idea which can be used to improve all such kind of GAN models,the authors try two models,GAN-CLS and AttnGAN_(GPT).As a result,in both of the two models,the proposed modified algorithm is more stable and can generate images which are more plausible than the original algorithm.Also,some of the generated images match the input texts better,and the proposed modified algorithm has better performance on the quantitative indicators including FID and Inception Score.Finally,the authors propose some future application prospect of the modification idea,especially in the area of large language models.展开更多
基金supported by the National Natural Science Foundation of China under Grant No.12288201。
文摘Synthesizing images or texts automatically becomes a useful research area in the artificial intelligence nowadays.Generative adversarial networks(GANs),proposed by Goodfellow,et al.in 2014,make this task to be done more efficiently by using deep neural networks(DNNs).The authors consider generating corresponding images from a single-sentence input text description using a GAN.Specifically,the authors analyze the GAN-CLS algorithm,which is a kind of advanced method of GAN proposed by Reed,et al.in 2016.In this paper the authors show the theoretical problem with this algorithm and correct it by modifying the objective function of the model.Experiments are performed on the Oxford-102 dataset and the CUB dataset to support the theoretical results.Since the proposed modification can be seen as an idea which can be used to improve all such kind of GAN models,the authors try two models,GAN-CLS and AttnGAN_(GPT).As a result,in both of the two models,the proposed modified algorithm is more stable and can generate images which are more plausible than the original algorithm.Also,some of the generated images match the input texts better,and the proposed modified algorithm has better performance on the quantitative indicators including FID and Inception Score.Finally,the authors propose some future application prospect of the modification idea,especially in the area of large language models.
