One of the latest buzzwords in technology and editing was AI. While the earliest implementations were a bit gimmicky, powerful tools and developments from companies like Adobe, NVIDIA and Luminar brought AI to the end user in a meaningful way. Photoshop’s March 2021 update introduces a new AI tool that promises huge resolution enhancements for any camera. Does it meet the hype?
If you’ve not caught up with Photoshop’s notes, Adobe Camera Raw has got a new feature called Super Resolution. Super Resolution is currently available in Camera Raw 13.2 and will soon be coming to Lightroom and Lightroom Classic, using a machine learning model to “enlarge photos intelligently while maintaining clean edges and maintaining important details.” In practice, this is a one-click way to double the megapixel count of your photo while retaining much more detail than the “dumb” upscaling, as with the two-and-closest methods.
Super resolution of Photoshop versus Topaz Gigapixel AI
However, this is not a new idea. AI scaling has been around for a while. It has even been implemented in existing consumer products, especially in Topaz’s Gigapixel AI software. Gigapixel is based on a similar principle: practice a machine learning model with a set of low and high resolution image pairs so that the computer eventually learns what a low resolution area can look like in high resolution. This model can then take the photos of an example and “create” details to fill in the blanks. Depending on how it is implemented, it can be quite intense, depending on the computer’s GPU to do a lot of work.
Although there are some differences between the operation of these programs, with PS’s feature integrated in ACR instead of a standalone program, and Gigapixel offers a few more options to customize the processing, but the end results are completely comparable .
For these tests, I wanted to look at a few different types of images that I take regularly and that often benefit from more resolution. For comparison, I grabbed some raw files from my Mavic Air 2 (to represent aerial photos) and my Nikon Z 7 (which represents architecture and product photography, as well as a higher resolution). Although these files were not processed, each program handled them slightly differently. The most important distinction is that ACR automatically applies lens corrections. This resulted in a slight difference between FOV and brightness between the files, but I do not consider it really relevant in the comparison, as you can send a processed file through Gigapixel without a significant difference. In the following images, the Photoshop Super Resolution version is on the left, with Gigapixel on the right.
Increasing resolution of drone photos
This is in my opinion the worst case scenario for scaling up. Although the Mavic Air 2 files are very impressive for a camera that can fly at 40 km / h, they are not beautiful at pixel level. They can make a bit of noise, even at low ISOs, and the Quad Bayer sensor, like Fuji’s X-Trans, has historically had issues with some demo processes.
One thing that pissed me off when I read about Super Resolution was how it included ACR’s step of reinforcing details by default. Enhance Details was an earlier battle with ML-powered tools and provides a way to unmask raw files with fewer artifacts. This is a very slight improvement in many cases, but I have found that it can help in difficult moire, or with atypical sensor setups like X-Trans or Quad Bayer. As a result, I do not do it by default, but realize that it is available.
This combination of improved image quality and increased resolution makes Super Resolution look like a very promising option for use with drones, and I can say it really delivers.
Let’s first talk about processing time and workflow. If you are loading the raw file in Photoshop, right-click on the image and select Enhance, the relevant menu will appear. From here, a preview is quickly generated and a new DNG can be created. Via this flow, you still have access to the same features you would have if you processed the raw file normally and you can quickly see what benefit Super Resolution offers.
With Gigapixel, loading the raw file and setting things up is quite a bit slower. There is a delay as the preview is generated, a significant delay each time you move or change an option when it is drawn again, and finally a very important difference in actual processing times. Super Resolution delivered a completed file in 3 seconds, while Gigapixel AI took 1 minute and 23 seconds.
In terms of finished files, the Photoshop version is significantly better. Two major improvements are visible. The first is an area that has been a problem for many other software tools for handling Quad Bayer or X-Trans files: ‘worm-like’ green areas. In the Gigapixel version, there is a very natural, unnatural appearance of this area of leaves.
The second major improvement is the relative absence of important artifacts in the Photoshop version. To personalize this, Gigapixel is too aggressive in detailing. It creates faint patterns in areas that should have a plain texture and generates striking artifacts in areas such as text and faces. Meanwhile, Photoshop just seems to be delivering a very good upscale. The drone shot becomes a 48-megapixel shot after processing. Although it does not match a DSLR for microcontrast and sharpness, it is surprisingly close and a drastic improvement over the original 12-megapixel recording.
The best option for scaling up architectural images
While my Z 7 offers excellent resolution with its 45-megapixel sensor, more is always better. Therefore, I was curious how these two scaling methods would work with a file that provides a mixture of organic shapes and straight lines, as well as fine details.
From this test file, I observed a similar pattern in usability, but to an even greater extent. Photoshop delivered a complete file in 6 seconds, while Gigapixel took 5 minutes and 1 second to complete the version.
Comparing the two files, Photoshop again produced a surprisingly neutral file. There are no major problem areas, and the files still have a bit of a “bite” at the pixel level. Since Photoshop applies lens corrections automatically, the FOV is a bit different, but I think these corrections should be applied to the Gigapixel file anyway, because there is noticeable distortion in the buildings. At the pixel level, PS’s version has only a minor issue with some fine details, such as the stars on the flags. In Photoshop, they are rendered as stars, but with a bit of false color creeping into them. In Gigapixel’s version, these are unrecognizable stains as well as artifacts of false color.
Gigapixel also runs down the watercolor problem along the chain link fence. Here, Photoshop makes the fence as expected, while Gigapixel’s version is greasy, and it seems like individual threads of the fence are almost out of focus.
In the architectural details, both are competent. Photoshop seems to be failing to maintain a bit more noise and texture, while Gigapixel makes things smoother to a greater extent, but I think you can move any file to the same place with a bit of sharpening and noise reduction.
Closure
For about $ 100, I just can’t see the value of Topaz’s Gigapixel AI product for my workflow, since Adobe’s Super Resolution is available. During my testing of the variety of subjects I shoot, Super Resolution yielded equal or better results anyway. Architecture, landscapes, night views, product photos, aerial photos and more all worked out better in Super Resolution. It does not even consider the important workflow benefits: Super Resolution is built into Photoshop, respects the existing ACR workflow better, and is 20 to 50 times faster to process. If you have not tried Super Resolution yet, definitely try it!