d1e39c9bda5c80ac3d8ea9d658163967-AuthorFeedback.pdf

Thank you for the helpful comments. Note that for our approach we don't perform an exhaustive search for We agree we leverage existing techniques (L166-8). In this work, we've integrated these techniques in a coherent framework to address an We'd like to point out that seminal works in learning disentangled rep-25 R3 states "..model tends to mode collapse on YTF" - we re-27 Faithfully modeling such image distribution hence results in similar generated images. Due to time constraints, we're only able to compare Uniform-InfoGAN and our final method. 's role is to push the discovered factor to better correspond with object Hence, Gumbel-softmax alone shouldn't be thought of as an Finally, we'll discuss the mentioned related works; we thank R2 for pointing them out.