One of the latest buzzwords in technology and editing has been AI. Although the first implementations were a trick, powerful tools and developments from companies like Adobe, NVIDIA and Luminar brought AI to the end user in a significant way. The March 2021 update to Photoshop features a new AI tool that promises massive resolution improvements for any camera. Does this live up to the hype?
If you are not up to date with Photoshop notes, Adobe Camera Raw has received a new feature called Super Resolution. Currently available on Camera Raw 13.2 and soon on Lightroom and Lightroom Classic, Super Resolution uses a machine learning model to “intelligently magnify photos, keeping edges clean and preserving important details”. In practice, it’s a one-click way to quadruple your photo’s megapixel count, while retaining much more detail than “dumb” scaling, such as the bicubic and closest neighbor methods.
Photoshop Super Resolution Versus Topaz Gigapixel AI
This is not a new idea, however. AI upscaling has been around for some time. It has even been implemented in existing consumer products, mainly in Topaz’s Gigapixel AI software. The gigapixel is built around a similar principle: training 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 might look like in high resolution. This model can then up-sample the photos and “create” details to fill in the blanks. Depending on how this is implemented, it can be quite computationally intensive, depending on your computer’s GPU to do a lot of work.
While there are some differences between how these programs work, with the PS feature integrated into the ACR instead of being a standalone program and Gigapixel offering more options for customizing the processing, the end results are perfectly comparable.
For these tests, I wanted to take a look at a few different types of images that I shoot frequently and that usually benefit from more resolution. For comparison, I took some raw files from my Mavic Air 2 (to represent aerial photos) and my Nikon Z 7 (representing architecture and product photography, as well as higher resolution). Although these files are not processed, each program handles them a little differently, the most important distinction being ACR applying lens corrections automatically. This resulted in a small difference between FOV and brightness between the files, but I’m not considering it relevant in the comparison, since you could pass a file processed by Gigapixel without a significant difference. In addition, in the following images, the Photoshop Super Resolution version will be on the left, with Gigapixel on the right.
Increased resolution of drone photos
In my opinion, this is the worst scenario for scaling up. While the Mavic Air 2 files are very impressive for a camera that can fly at 40 mph, they are not beautiful at a pixel level. They can be a bit noisy even at low ISOs, and the Quad Bayer sensor, like Fuji’s X-Trans, has historically had problems with some demosaicing processes.
One thing that caught my attention when reading about Super Resolution was how it included the enhanced detail processing step of ACR by default. Enhance Details was a previous foray into ML-based tools and offered a way to demosaic raw files with less resulting artifacts. It is a very small improvement in many cases, but I have found that it can help in cases of complicated moiré or with atypical sensor configurations like X-Trans or Quad Bayer. As a result, I don’t do this by default, but I appreciate it being 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 that it really does.
First, let’s talk about processing time and workflow. Load the raw file in Photoshop, right-click on the image and select Enhance opens the relevant menu. From here, a visualization is generated quickly and a new DNG can be created. Through this flow, you still have access to the same resources you would have if you were processing the raw file normally and you can also quickly see the benefit that Super Resolution will offer.
With Gigapixel, loading the raw file and setting things up is a little bit slower. There is a delay as the preview is generated, a significant delay each time you scroll or change an option as it is redesigned, and finally, a very significant difference in the actual processing times. Super Resolution produced a finished file in 3 seconds, while Gigapixel AI took 1 minute and 23 seconds.
As for the 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 when dealing with Quad Bayer or X-Trans files: green areas with “worms” appearance. In the Gigapixel version, there is a very watercolor and unnatural appearance in this area of foliage.
The second major improvement is the relative absence of important artifacts in the Photoshop version. To personify this, Gigapixel is overly aggressive in “inventing” details. It creates faint patterns in areas that should have a simple texture and generates noticeable artifacts in areas such as text and faces. Photoshop, on the other hand, seems to offer only a good upscale. The image of the drone, after processing, becomes a 48-megapixel image. Although it will not match a DSLR in microcontrast and sharpness, it is surprisingly close and a dramatic improvement over the original 12 megapiel photo.
The best option to increase the scale of architectural images
Although my Z 7 offers great resolution with its 45 megapixel sensor, more is always better. For that, I was curious to know how these two sizing methods would work with a file that offers a mixture of organic shapes and straight lines, along with some fine details.
From that test file, I observed a similar pattern of usability, but to an even greater degree. Photoshop rendered a finished file in 6 seconds, while Gigapixel took 5 minutes and 1 second to finish its version.
Comparing the two files, Photoshop again delivered 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 automatically applies lens corrections, FOV is a little different, but I think these corrections would need to be applied to the Gigapixel file anyway, as there is noticeable distortion present in the buildings. At the pixel level, the PS version has only a small problem with some fine details, like the stars on the flags. Photoshop renders them like stars, but with a little bit of fake color showing up. In the Gigapixel version, these are unrecognizable spots, as well as false color artifacts.
Gigapixel is also faced with that watercolor problem again along the wire fence. Here, Photoshop renders the fence as expected, while the Gigapixel version is blurred, with individual strands of the fence almost seeming to blur out of focus.
In architectural details, both are competent. Photoshop seems to be on the wrong side of preserving a little more noise and texture, while Gigapixel smooths things out to a greater extent, but I think you could push any file to the same place with a little bit of sharpness and noise reduction.
Conclusion
For about $ 100, I just can’t see the value of Topaz’s Gigapixel AI product for my workflow now that Adobe Super Resolution is available. In my tests on the whole range of subjects I photograph, Super Resolution showed equal or better results in all cases. Architecture, landscapes, night landscapes, product photos, aerial photos and much more are better in Super Resolution. This without considering the significant benefits of the workflow: Super Resolution is integrated with Photoshop, better respects the existing ACR workflow and is 20 to 50 times faster to process. If you haven’t tried Super Resolution yet, give it a try!