September 29, 2021

ML Super Resolution gets an update

Our machine learning models practically live a life of their own. For most of our apps, features and improvements have to be worked on for each app separately. But when it comes to ML models, we can include improvements everywhere at once. And that’s what has just happened with ML Super Resolution – the algorithm has been refined to work much better with illustrations and graphics, especially those that include semitransparent areas – and those changes are available in both Pixelmator Pro and Pixelmator Photo as of their most recent updates.

Our ML Super Resolution algorithm is pretty unique because it supports transparent images and this adds an extra level of complexity to the algorithm. In images with transparency, upscaled edges weren’t quite as sharp as we’d like. So we made a few tweaks and have some pretty big improvements that we’re very proud of! Check them out below.

Old ML Super Resolution version

New ML Super Resolution version

And in other ML news, the sixth-generation iPad mini is now available. It sports the new A15 Bionic chip with a 16-core neural processor, which is around 40% faster than the previous generation. We ran a quick test that we’ve done previously and the iPad mini came in at 0.41s to upscale a 0.3 megapixel image. That’s faster than the 2017 iMac Pro!

Mac
Compute time*

MacBook Pro (13-inch, Mid 2012)
61.6s

MacBook Air (13-inch, 2018)
13.7s

iMac Pro (2017)
0.56s

M1 MacBook Air (2020)
0.51s

iPad mini (Sixth generation)
0.41s

* For this test, a 300,000 pixel image was upscaled to three times its original size. The iPad mini was tested on iOS 15, the M1 Mac on macOS 11, and other devices on macOS 10.15.