Old Photo Restoration via Deep Latent Space Translation (https://paperswithcode.com/paper/old-photo-restoration-via-deep-latent-space)

In the article, it says : "We propose to restore old photos that suffer from severe degradation through a deep learning approach. Unlike conventional restoration tasks that can be solved through supervised learning, the degradation in real photos is complex and the domain gap between synthetic images and real old photos makes the network fail to generalize ... "

I have a bit of difficulties to understand how they built their dataset of photos. At the beginning, I thought the idea was to create synthetic old photos from a set modern photos like IMDB-WIKI – 500k+ face images with age and gender labels. Hence, I thought the idea was to use deep learning with the synthetic photos dataset (artificially degraded) as input and the modern photos as target. Once the model is train, we could've use it with real old photos instead of artificially build photos. It seems not the way to go. Can you explain to me how they generate their dataset and use it in their VAEs?

  • 1
    $\begingroup$ You're not really quoting the part of the paper that is relevant to answer your question. They explain how they build the datasets in section 4 of the same paper. Also, please, focus on one question. It seems that the VAE part is another different question than the question "how they build the dataset" $\endgroup$
    – nbro
    Oct 18 '20 at 18:40
  • $\begingroup$ @nbro Thanks for that! You are right! I overlooked this part. Do you think it is hard to generate the training dataset from that section. $\endgroup$
    – David
    Oct 18 '20 at 18:46
  • 1
    $\begingroup$ I didn't look at the details. I just read that they use Gaussian noise and stuff like that. Gaussian noise shouldn't be expensive, i.e. you just need to sample from a Gaussian distribution. Maybe that could be a separate question, but please focus on a very specific problem/question for each post. $\endgroup$
    – nbro
    Oct 18 '20 at 18:55

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.