Google is using advanced AI to compress images that are smaller than JPEG files, and they’re doing a good job at it.
By now it doesn’t seem like it’s such a big feat, because one can actually meet decent compression rates for web without much loss of quality. But imagine having an image that is big and clear, but the loss in quality isn’t really noticeable, clubbed with a smaller space size, and that’s where Google wants to aim.
Google is making use of AI to reduce the size of image files.
So this is how Google did it. They first took a random sampling of over six million images with a resolution of 1280×720 each from the web. Then they broke those images down, into 32×32 tiles, and then picked the 100 worst compressed images. So by first working on the hardest images, it would then be easier to finish improving the rest. Google sure knows what they’re doing!
Now here’s the magical part of the whole process – how they actually got to compress the images without losing quality. Google open-sourced a very interesting tool last year called TensorFlow. And what TenserFlow does is, it learns how complicated neural networks within system architectures work and adapts from that learning. Simply put, TenserFlow was put to the task of learning the different compression methods those images went through and adapted to create a compression method to provide better results. Sounds like Sci-Fi stuff, doesn’t it?
Well, it worked. And the new models which the researchers collected, thanks to TensorFlow helped them beat JPEG’s standard compression method. But researchers aren’t fully sure whether the end result will be what the user will want to see as the human perception is vastly different from the technicalities and numbers, so there isn’t really a clear winner. Beauty is in the eye of the beholder after all. But this is a step towards the right direction.
They are hoping to next try compressing video files. Open sourced AI tools such as TenserFlow are an effective means to derive solutions to complex problems and having it accessible to the masses, means more innovation and efficiency for us all. A win-win situation!
Now until TenserFlow doesn’t learn how to somehow take over the world, we’re all safe.