HN Explorer

Ask HN: Neural Nets for lossy image compression?

1 points by AlexanderTheGr8 ↗ HN
Lossy image compression basically means compressing limited information for human eye to perceive the essentials of the image.

So we skip information which human eye cannot perceive and which is redundant (i.e. which is not necessary to understand the essentials of the image).

Neural Nets are amazing at figuring out what human eyes can understand. Can't they be used for some amazing lossy image compression?

IMO, No hard-coded algorithm can work as well as a neural net to figure out how much information is needed for the human eye to understand an image.

What do you think? Also are there any discord/other channels to discuss data compression?

6 comments

[ 9.3 ms ] story [ 36.3 ms ] thread
Skyy93
What you search for is an autoencoder.
ipunchghosts
In practice these rarely work for useful image compression.
AlexanderTheGr8
I have already seen research for using GAN for image compression. I was primarily looking for some discussion groups, considering that compression in today's big-data world is increasingly becoming one of the most important unsolved-problems.
ipunchghosts
See works by Johan balle and Lucas this. Lots of work done by them on this topic.
AlexanderTheGr8
Thanks a lot for the suggestions. Interestingly I have noticed a lot of image-compression research coming out of Google (and no other big-tech company). I wonder why.

Also, do you happen to know of any discussion groups for image compression?

Skyy93
I am sorry, I do not know good discussion groups about image compression. If you are a researcher you can try to reach some of your colleagues that are into this topic. Perhaps they knew a good source/discussion platform?