This submission received borderline reviews. After the post-rebuttal discussion, the most negative reviewer (who also provided a very cursory review) has come around slightly and is willing to increase their score to a borderline accept (though they have not actually done so in CMT). One of the main concerns was similarity to a piece of prior work (Meshry et al.) However, the rebuttal seems to have sufficiently addressed this--though the methodology is similar, the application is significantly different. Another concern is that the performance of the BiGAN appears quite poor. None of the reviewers seem to have enough expertise to judge whether the model just wasn't trained appropriately (as these models are very tricky to train even for experts), or whether this is a more fundamental problem. Nevertheless, the paper's main contributions don't seem to depend critically on this, so it seems not to be a clear reason to reject the paper.