Learning Convolutional Features and Text Information to Draw Image

In this paper, a more effective and general joint exploration method (JEM) is proposed to synthesize images. By combining the technology of image segmentation, feature extraction, and image synthesis, high-quality images can be generated based on the text description and the convolutional segmentation information. Experiments on the Oxford-102 dataset show that our method is more effective than the GAN-CLS-INT method proposed recently. It also shows that in the training process, using VGG for feature extraction has a faster convergence speed than using AlexNet. Simultaneously, we demonstrate that the segmentation image's background information plays an active role in the training process.