![]() ![]() So it's six of one, half a dozen of the other.For most photographers, the digital darkroom is based around payware or subscription software from the likes of Adobe, Capture One, DxO or one of their many rivals. I therefore find that results from the above procedure are roughly equivalent to what I would get by doing a 'straight' conversion using a better demosaicing and sharpening engine (I've been using Rawtherapee). Having said that, I personally find that a fair bit of detail and acutance is lost early on, as a result of Adobe's demosaicing engine being less than optimal in terms of detail interpretation. The idea of having AI-invented details in my images fills me with horror, as I'm sure it does many photographers, but used in this way, it doesn't feel much different to any other post-processing technique. So I'm essentially thinking about it as a sophisticated sharpening tool, and viewed as such, it works pretty well. However, if I resample the result down to the original size, stack it above the straight conversion and set the transparency to around 50%, the artifice is hidden sufficiently, whilst leaving some improvement to the apparent acutance of details over standard sharpening techniques. Generally, I find that its 'naked' output is not acceptable 'as is' - just too many artificial-looking details and textures. I've been doing quite a bit of raw conversion recently, and being trying out Adobe super res as I go along. It's good for hard edges and suchlike, and makes up stuff in the small details that I didn't like much. This is consistent with what I saw in other pictures. You find that it invents contrasty little things in the roof tiles that weren't there whereas when you look at those vases or the white streaks on the chimney, not much is gained vs conventional upscaling. What you gain with the stacking method is not so much in resolution but in reducing artifacts (oh I cut off the part with the moiré, but yo can see a hint of this around the cables).ĪCR superres looks impressive - way better than conventional upscaling - until you compare it with the picture from the 85mm lens, or what I would call Reality as opposed to The World Invented By An Algorithm. When you look at the comparison at 100% or more, you'll notice: I ran the raw file from a) through Camera Raw's new superresolution feature (which took ages) and, again, scaled it down to fit the picture taken with the 85mm lens (85%)Īnd finally took a picture with the 85mm lens (Z-Nikkor at f:5,6 - I didn't compensate for the diminishing dof but that hardly matters here) and did not resize. I took another 16 pictures and combined them into a superresolution file according to, then scaled it down to fit the picture taken with the 85mm lens. ![]() I took a picture with the 50mm lens (Nikon Z7, 50mm 1,8 S Nikkor at f:5,6, you see a crop from the center), upscaled it to 170% (so that the crop had as many pixels as the same crop when taken with the 85mm lens), applied unsharp mask etc - that's about as good as it gets, conventionally. I wanted to know whether theses pixels are any good and, on a slow business day, did this: Adobe claims that this doubles linear resolution, or gives you four times the pixels. So I tried this new feature in ACR, machine-learning based superresolution. ![]()
0 Comments
Leave a Reply. |