I was on vacation last week, so this newsletter is going out a few days after the latest Smarter Image column appeared. Computational photography fascinates me in how machine learning is being applied at several different levels of the image-making process. One of those is at the root level of translating the data in a raw file, before you start working with tone and color and the rest of the adjustments we think of when editing an image.
In the new column, I look at DxO PureRAW 2, which handles the demosaicing and denoising stage. By running a raw file through PureRAW before editing, you can start with a cleaner image, particularly if the original is noisy. I found that it also creates good results for FujiFilm .RAF files, translating Fuji’s X-Trans sensor data better than other tools such as Lightroom and Capture One.
A couple of readers have already written to say they don’t like how PureRAW handled the night sky image, specifically the white halo around the edges of the tree. I agree, and I’m not sure if it’s because of that image, or the file format (Sony), or something else. But I do like how it handles the noise and the tones within the tree’s trunk.
Read the piece here: Preprocess Raw files with machine learning for cleaner-looking photos.