With conditional GANs it is possible to generate images of a certain class of objects. And moreover with current text-to-image methods it seems to be possible to control certain details of the generated images as well, as i.e. this research shows.

In my case I analyze a data set of labeled car images where the labels contain ca. 20-30 features such as brand, color, shape, type, etc. Currently I have two questions concerning this problem:

  1. Is it possible to train a GAN to generate realistic images with a resolution of less than 256 by 256 pixels of cars?
  2. And will the GAN generalize to unseen feature vectors?

Generally speaking: Given a feature vector of around 20-30 categorical variables controlling details of the image. Is it possible with state-of-the-art methods to build a GAN-generator that takes a feature vector as input and generates an image with the details specified in the feature vector?

  • $\begingroup$ could you please add full reference for your link in case it dies in the future? Thanks! $\endgroup$ – Antoine May 6 at 9:28

it is possible .But not in that way. you need to offer randomly sampled input e.g. from uniform distribution and let the GAN learn the real data distribution it self but not design the input manually first.

| cite | improve this answer | |
  • $\begingroup$ Hi, welcome to CV! We are looking for longer and well justified answers (with references, examples, figures, etc.) As it currently stands, your answer may be flagged as a comment and removed. $\endgroup$ – Antoine May 5 at 8:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.