Adobe has introduced a new feature in Camera Raw 13.2. They call it Super Resolution. I’ll call it SR here.
SR is a demosaicing technique that produces an output image with twice the number of pixels as the raw file in both dimensions. That means the file as four times the number of pixels of the raw file. SR is AI-based, and builds on the Enhance Details feature that we’ve had for a year or so.
I tested it with a Siemens Star, which is probably playing to its strengths. I’m guessing such a star was part of the training set.
The scene, with a GFX 100 and the Fuji 110 mm f/2 lens at f/2, Lr default settings except that sharpening amount set to 20 instead of the default 40. I white balanced to the gray parts of the star afterwards, and the balance isn’t quite the same in the crops below. I apologize for that.
Three closeups, at 1:1 magnification. Click on the image, then tell your browser to use 100% magnification.
Good performance. False color essentially eliminated with both of the AI demasicing techniques.
Zé De Boni says
Impressive! First thing that came to my mind was what will we get when we combine SR with Pixel Shift Multishot. I don’t mean you should do this test, but just wonder how it would be to work on a 1.6 Gp file (~1.0 for my A7R4). The industry should provide us proxi files as is done for video, which could be worked in RAW converters and image editors. That is already necessary as we are dealing with tens or hundreds of the now standard high resolution images.
algrove says
I wonder how this compares to a like Topaz gigapixel AI enhancement.
JimK says
I didn’t do that in combination with Enhance Details, but here’s how it looks with regular Lightroom demosaicing.
https://blog.kasson.com/nikon-z6-7/megapixels-smart-resizing-and-printing-topaz-gigapixel-ai/
Christian says
beside the amazing NR i see similar effects of artefact suppression and better detail with dxo deep prime too when used on low iso shots. seems ai based decoding sets a new standard. it will become a challenge for other companies especially c1 to compete in this field.
@Zé De Boni the idea to work on proxies files was already available 25 years ago in the at the time ground breaking livepicture software but sadly the much inferior photoshop won the game.
JimK says
Internally, Lightroom uses a technique similar to Live Picture.
Brandon says
Most people seem to think of SR as a way to make a low MP camera perform closer to a high MP camera and that it is not as useful for high MP source images.
I find that the opposite is true. The more megapixels you feed it, the more synthetic data it produces. SR on a 25 MP image produces 75MP of synthetic data whereas the same on a 100 MP image produces a whopping 300 MP of fake data. The amount of artificial information added to the 100mp image is 12x more than the 25MP image even started with.
Adobe should really open up the megapixel cap they put on that feature because the added detail is surprisingly useful for panoramas and gigapixel sources.
What’s interesting is that it can allow for resolving beyond the Rayleigh limit in a surprisingly “real” way (like in your Siemens Star example). I gained a lot of respect for the feature when it got the lettering on a clearly illegible road sign correct, achieved pixel perfect enhancement of fine lines like power lines and stray hairs, and it properly rendered details in various textures like roofing shingles and bricks.
The real test is SR at f/16 vs regular image at f/11. It’s a synthetic battle against the aberration limit. It makes your camera feel like it is resolving with a larger aperature.