We propose a novel method for solving regression tasks using few-shot or weaksupervision . At the core of our method is the fundamental observation that GANsare incredibly successful at encoding semantic information within their latentspace, even in a completely unsupervised setting . By leveraging this observation, our method turns a pre-trained GAN into aregression model, using as few as two labeled samples . This enables solvingregression tasks on datasets and attributes which are difficult to producequality supervision for . We show that the same latent-distancescan be used to sort collections of images by the strength of given attributes,even in the absence of explicit supervision . Extensive experimental evaluationsdemonstrate that our method can be applied across a wide range of domains, can be used across a number of domains . It can achievestate-of-the-art results in a few-hunter-hunter settings, even when compared to methods designed to tackle a single task

Author(s) : Yotam Nitzan, Rinon Gal, Ofir Brenner, Daniel Cohen-Or

Links : PDF - Abstract

Code :
Coursera

Keywords : method - supervision - attributes - latent - gan -

Leave a Reply

Your email address will not be published. Required fields are marked *