Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Neural Enhance – Super Resolution for images using deep learning

github.com

Post reply on HN
Print view View original

1–10 of 53 posts

Re: Neural Enhance – Super Resolution for images using deep learning

#3

Post by anilgulecha »

This is amazing. The surprise is that while the higher resolution images seem real, they are reconstructions based on the previous learning, and can be very different from the actual.

Nice test images to include would have been an original image, downsampled image, and the reconstructed image. If the author is reading this, could they add this to the README?

Sci-fi on TV is making it to the real world :)

Re: Neural Enhance – Super Resolution for images using deep learning

#4

Post by dharma1 »

nullc wrote: post #2

Comparison using nearest neighbor, instead of a more reasonable linear filter, or-- heaven forbid-- some edge basic directed interpolator... is a little cheaty.

Agreed, it would have been nice to show other upscaling algorithms. But neural net super resolution generators can still have significantly more detail at 4-8x, as shown here http://arxiv.org/abs/1609.04802

Re: Neural Enhance – Super Resolution for images using deep learning

#5

Post by ehsanu1 »

anilgulecha wrote: post #3

This is amazing. The surprise is that while the higher resolution images seem real, they are reconstructions based on the previous learning, and can be very different from the actual. Nice test images to include would have been an original image, downsampled image, and the reconstructed image. If the author is reading this, could they add this to the README? Sci-fi on TV is making it to the real world :)

Would also be interesting to see some pixel art run through this. It probably won't work that well given that its trained on real downsampled photos though, but who knows.

Re: Neural Enhance – Super Resolution for images using deep learning

#9

Post by fezz »

anilgulecha wrote: post #3

This is amazing. The surprise is that while the higher resolution images seem real, they are reconstructions based on the previous learning, and can be very different from the actual. Nice test images to include would have been an original image, downsampled image, and the reconstructed image. If the author is reading this, could they add this to the README? Sci-fi on TV is making it to the real world :)

The examples show a comparison to the original and the scaled.
Post reply on HN