Data collection of toy horses and zebras for multi-modal image synthesis

Motivated by building a live demonstration of horse-to-zebra synthesis, we created our own dataset consisting of images of different toy horses and zebras. We used a wide variety of backgrounds, camera angles and lighting settings. In addition, images were also created in which the toy animals were held in the hand to better demonstrate synthesis in front of the camera. Besides the raw image and video footage, we prepared different compilations of data in form of datasets. Following the horse-to-zebra synthesis by Zhu et al. in the paper "Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks", we partially adopted the horse2zebra dataset from real horses and zebras in our datasets.为搭建马到斑马图像转换的实时演示系统，我们构建了专属数据集，该数据集收录了不同玩具马与玩具斑马的图像。我们采用了多样的背景环境、拍摄角度与光照设置；此外，我们还拍摄了手持玩具动物的图像，以更好地在镜头前展示图像转换效果。除原始图像与视频素材外，我们还以数据集的形式整理了多组数据合集。参考朱等人在论文《使用循环一致性对抗网络的非配对图像到图像转换》（Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks）中提出的相关工作，我们在自建数据集中部分采用了基于真实马与斑马的马到斑马数据集（horse2zebra）。