Zero-shot blind learning for single-image super-resolution

HIGHLIGHTS

SUMMARY

    Most existing deep learning methods are usually realized in a fully supervised way, and then, the learned SR model may only be applicable to the LR observations captured in controlled imaging conditions such as with the assumed degradation 2 of 13 model: "bicubic" downsampling. To handle the limitation of deeply relying on the prior training with external data, internal learning via extracting training samples from the observed LR image and its downsampled version has been exploited to produce a specific SR model for the understudied image. To solve the challenge of the blind SR . . .

     

    Logo ScioWire Beta black

    If you want to have access to all the content you need to log in!

    Thanks :)

    If you don't have an account, you can create one here.

     

Scroll to Top

Add A Knowledge Base Question !

+ = Verify Human or Spambot ?