I just took a course on deep learning where one part of the syllabus was image classification and object recognition using CNNs, but I wonder how deep learning can be applied to apply certain filters or complex edition on a given photography. For example, a common action on basic photography is to remove light reflexion from human eyes when doing a portrait photo.

How could this be done with a deep learning algorithm? The simplest idea seems to train a network using two versions for each photography in the training set: one without light correction and another one with it. But I guess that there might be more suitable procedures to perform such kind of actions with better results.

I would love to hear what the community has to say about this. Any recommendation?

  • $\begingroup$ what you want is called style transfer. there are a couple of approaches. the most promising are based on GANs. Pix2Pix uses two pairs of images and learns a mapping from one domain to anther, CycleGAN does the same thing but doesn't require paired samples. there are also StyleGAN and StarGAN $\endgroup$ – Aray Karjauv Dec 7 '20 at 21:15

Your Answer

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

Browse other questions tagged or ask your own question.