Transforming and Projecting Images into Class-conditional Generative Networks

We present a method for projecting an input image into the space of a\nclass-conditional generative neural network. We propose a method that optimizes\nfor transformation to counteract the model biases in generative neural\nnetworks. Specifically, we demonstrate that one can solve for image\ntranslation, scale, and global color transformation, during the projection\noptimization to address the object-center bias and color bias of a Generative\nAdversarial Network. This projection process poses a difficult optimization\nproblem, and purely gradient-based optimizations fail to find good solutions.\nWe describe a hybrid optimization strategy that finds good projections by\nestimating transformations and class parameters. We show the effectiveness of\nour method on real images and further demonstrate how the corresponding\nprojections lead to better editability of these images.\n