基于pix2pixHD的行人图像生成

Pedestrian Image Generation Based on Pix2pixHD

本文提出了一个基于行人姿态mask利用条件生成对抗网络来生成逼真的行人图像的方法,使用了一种最新的adversarial loss和最新的多尺度生成器-判别器结构来增加提高行人图像细节。实验表明,通过使用成对的行人姿态mask和行人图像训练后的pix2pixHD网络可以生成了一幅视觉吸引人的具有目标姿态不同属性性质的行人图像。

This paper proposes a method based on the pedestrian pose mask to generate a realistic pedestrian image using conditional generation confrontation networks.In this paper,we use a new adversarial loss and the latest multi-scale generator-discriminator structure to increase the detail of pedestrian images.Experiments show that by using the paired pedestrian pose mask and the pix2pixHD network trained by the pedestrian image,a visually attractive pedestrian image with different attribute properties of the target pose can be generated.