Unsupervised SIFT features-to-Image Translation using CycleGAN

The generation of video content from a small set data representing the features objects has very promising application prospects. This is particularly important in context work MPEG Video Coding for Machine group, where various efforts are being undertaken related to efficient image coding machines and humans. representation feature points well understood by form, which easy understand humans, an current challenge. paper presents results on ability generate images SIFT without descriptors using generative adversarial network CycleGAN. impact keypoint method learning quality presented. subjective evaluation generated