A couple days ago, Adobe released Lightroom 3.0 Beta 1, which introduces a variety of new features, but also introduces a new RAW processing engine. While many of these features are fairly simple to appreciate, the changes to the RAW processing engine are just as exciting, and worth a look as well.
I will intentionally avoid talking about performance differences. While I haven’t noticed particular performance differences between LR2 and LR3B1, I haven’t looked and wouldn’t consider any such comparison valid until it was made on a “release version vs. release version” basis, the performance of beta versions often differ from that of release versions for a variety of reasons that I won’t belabor here.
But if history is any guide, we can start looking at image quality, particularly with promised improvements to noise reduction and sharpening. Be careful not to read too much into these results, Adobe has intentionally turned off luminance noise reduction in LR3B1. Since it’s off by default in LR2, that shouldn’t impair our comparisons too much, but it is something to keep in mind.
To compare the two, I looked at differences in the results of the raw processing engines on a few ISO 3200 images of a camera bracket from my Canon EOS 1Ds Mark III. I’ve included a pair of 100% crops of raw conversions from the same file for your comparison. These samples are illustrative of the sorts of changes I’ve seen in the handful of other files I’ve examined so far. Both samples were converted entirely “as shot” with default settings for each version of Lightroom.
Because I’ve had to convert the crops to JPEGs here it’s possible that I’ll see some things you don’t, or see them more closely, so let me describe to you what I observe. In the bracket itself, color noise is enormously reduced in the bracket and the background, it’s almost non-existent in the Lightroom 3 Beta 1 sample. As a result, I’d be hesitant to use the LR2 version of this image for many purposes that the LR3B1 version would suffice for.
The character of the remaining luminance noise in both images is different, with the LR3B1 noise having a smaller grain size, the LR2 version seems mushier along the brackets. For very high-resolution images I’d prefer the LR3B1 version: slightly-brighter but much finer-grained noise.
The sharp highlights of both images (the bottom stroke of the letter D) show some chromatic aberration. But that aberration (despite identical CA settings) is much more apparent in the LR2 version than in LR3B1. Even turning on defringing in LR2 didn’t reduce the artifacts on the top of that D to the level that LR3B1 manages with “defringe” set off.
The blue reflections in the longer strokes of the D are extremely telling, the LR2 versions are soft and muddy, the LR3B1 versions feel enormously cleaner with little or no haloing.
As I pan around each of these samples in LR at high zoom factors, I notice that the LR2 version shows an uncorrected version of the image while panning and flickers in a cleaned-up version when I stop panning; the LR3B1 experience is cleaner and less distracting.
In the Detail panel, LR3B1 adds an “edge detail’ slider to it’s sharpening palette. With only color NR on, very low amounts of “edge detail” remove thin color details (such as the blue “D” reflections) but give cleaner shapes to those reflections. High amounts of “edge detail” provide more color in those reflections but also produce a grainier shape. The default settings do a great job of preserving the best of both worlds. Low settings also beat down on CA/fringing artifacts more, although I’d expect that you’d want to use the CA controls to adjust those before attacking them with the edge detail control.
Finally, LR3B1 appears to fix the venerable greyscale conversion bug I mentioned a few months back, which was a real problem for a small number of my black and white images.
While it is always risky to say too much in a first look at a first beta of a new product, I am deeply impressed with these new results. Lightroom is getting even better. Congratulations, Adobe!
Jason Grubb
26 Oct 2009So excited for LR 3.0. Can’t wait to get in there and explore
-Jason
Ihor Vorotnov
5 Dec 2009I can tell that LR3Beta works definitely faster. I even used the same large catalog to make the comparison closer to real world. I guess the key thing in such comparison is you hardware configuration. I tested on my old PC with Pentium 4 3.0GHz and 1,5Gb DDR PC3200 memory. This hardware is almost ancient now (but it still works fine, anyway) but it’s good for this kind of benchmarking. The LR3Beta performance is really much better. It takes almost the same time to render 1:1 previews but it works 25% faster when working with spot removal tool and 50-75% faster when applying develop presets.
Joe Decker
6 Dec 2009@Ihor: Thanks for the info! Very happy to have a snappier spot removal–for some reason that tool has always felt sluggish and “not quite right” to me.
Ben
6 Jun 2010based on your images above, they dont look proper to me, the LR2 is under exposed and the LR3B is over exposed.
where you just sliding controls here and decide if it looks good to you OR are you doing this on a more quantifiable environment, where you can replicate it later if needed.
If these where the case, im not impress with LR3B at all, the image has a feel of overly digitized look.
Hope this is only a Beta issue and not for the official release but still, i believe you could have done better with your test.
thanks for sharing though.
Joe Decker
7 Jun 2010Ben: It is impossible to numerically reproduce equivalent results between the two processing engines, the “same settings” on the sliders have qualitatively different meanings between the revisions. It is unavoidable that the proper trade-off between noise, sharpening, and so forth be subjective in such cases, that is a weakness, but an unavoidable weakness, of this first glance.
However, I think that when I look at these images in terms of the physical prints they can produce, it is clear to me (on my calibrated monitor, based on years of experience making fine prints) that the LR3B image will produce a better result. This result is confirmed by months of work with LR3B1 and LR3B2 since this note was posted, but that’s not something I can “prove” to you on a web page, nor something you should take blindly on faith. It is merely my experience, an experience I offer to you for whatever it is worth to you.