“Zoom in…and ENHANCE”
February 20, 2019
If you’re a Fujifilm X shooter, you might have a bit of a love-hate relationship with your camera. I absolutely adore my X-Pro2. It’s everything I could possibly want from a digital camera, but its “revolutionary” X-Trans sensor can often leave a bit to be desired when paired with a third-party raw processor. Images from X-Trans sensors had a bit of a reputation for looking like smeary, mottled, oil-painting like shit in certain circumstances. This leaves Fuji X shooters getting great JPEGs from the camera, but with a RAF raw file that simply can’t match the proprietary processing.
I’m an Adobe Lightroom user (the real Lightroom, not the cloud-based experimental toy thing that’s now called “Lightroom CC”) and have seen X-Trans RAF support go from “oh my god that is terrible is there something wrong with the camera?” to “this is pretty…pretty good”. LR’s processing is good enough most of the time, but for critical work, I use a standalone processor — Iridient X-Transformer — to convert the RAF into DNG and go from there. Iridient’s software deals with small details and foliage particularly well (see below).
“Give me a hard copy right there”
But Adobe released a new version of Lightroom “Classic” CC recently which includes a curiously named feature — Enhance Details. At first I thought all my Blade Runner/CSI fantasises were coming true — my very own ESPER machine! I could FINALLY understand how Deckard found Zhora in a scanned photograph using the completely sensible voice commands of: “Enhance 224 to 176. Enhance, stop. Move in, stop. Pull out, track right, stop. Center in, pull back. Stop. Track 45 right.”
But no, apparently this isn’t Adobe’s goal…yet.
Instead “Enhance Details” harnesses the powers of machine learning to improve details and reduce artifacts in images. It works with all standard Bayer raw files (from almost every other manufacturer) and, importantly, raw files from X-Trans sensors. As LR is at pains to point out to the user at every step, Enhance Details is “COMPUTATIONALLY INTENSIVE” and works best with a beefy GPU. After it processes your image, it plonks a DNG file into your filmstrip, next to your existing file.
So I thought I’d give it a spin on the same RAF file as above and see how the grille which LR’s native processing struggled with fared harnessing the POWER OF ARTIFICIAL INTELLIGENCE.
First off, it’s quite a slow process. A very slow process, in fact. This image took about a minute to do and the fan noise on my Mac got so loud, I was worried it was about to explode in flames. Luckily, it did not. Turns out ENHANCE DETAILS worked pretty well. In fact, better than the Iridient-processed file in some respects. Certainly the grille is almost entirely free of artifacts in the enhanced image, though the differences are much less stark elsewhere.
Curious, I thought I’d try another image. This one had a bit of foliage and natural stuff in it, as well as one of my mini-monsters. It isn’t the most technically perfect of images, but when photographing kids, you don’t get the opportunity to keep snapping until you get it right. While in this image, there is a substantial difference between default LR processing and Iridient, there is less difference between Iridient and LR’s ENHANCED image. In fact, looking at the jeans, I’d say there’s less detail in the LR ENHANCED image than the Iridient-processed file.
Great X-Pectations
Adobe’s processing of X-Trans files has come along in leaps and bounds. This, in part, is due to Adobe and Fujifilm working together to improve the algorithms. But it still leaves a bit to be desired. ENHANCE DETAILS improves on LR’s X-Trans processing considerably, but lies somewhere between LR’s default processing and Iridient for final image quality.
For casual images, I’ll keep using LR’s own processing algorithms because, for the most part, they’re fine. For critical images, I’ll stick to Iridient X-Transformer, but I’ll also run a version through ENHANCE DETAILS to see how it looks. Being a machine learning-based tool, it no doubt can and will improve over time, but it’s not quite there yet.