Pages: [1]   Go Down

Author Topic: DXO OpticsPro 10 vs. ACR  (Read 11123 times)

Redcrown

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 431
DXO OpticsPro 10 vs. ACR
« on: November 29, 2014, 04:10:16 am »

I'm a long term ACR user, but every now and then I like to play with other raw converters. So I'm testing the new version of DXO OpticsPro 10. I know DXO OpticsPro has a few loyal fans, but I'm just not getting it. It seems to fail in every test I give it compared to ACR. So maybe it's operator error?

Below is a Dropbox link to a Canon 5D3 DNG file. I'd apperciate any DXO gurus working on this file and showing or explaining their steps. Show me if I'm missing something.

I've processed several of my old raw files, and found this one presents several common problems. So it's a good test image.

It has a wide dynamic range. There are deep shadows, but no clipping on the low end. There are bright highlights, with clipped green and blue channels in the clouds. The red channel has only a small clipped area. There are fine details in the dark trees, and being against the bright cloud causes significant chromatic abberation.

I'm not testing lens correction yet, and only basic noise reduction. Just focsing on basic conversion - white balance, color rendition, exposure correction, shadow and highlight recovery.

Problems I'm seeing: I can't get DXO to even come close to recovering the highlights in the sky and clouds as well as ACR does. With ACR I get detail in the blown out cloud and get some blue sky back. No so with DXO. And DXO's highlight recovery appears to reach into the midtones and shadows much more than ACR does, causing an unappealing flattening of tones in the foreground path.

ACR does a very good job removing chromatic abberation, while DXO seems to struggle. Plus, DXO apperars to generate some strange artifacts. Look at what happens to the tree line in the top right section.

https://dl.dropboxusercontent.com/u/62166185/ire038.dng
Logged

kirkt

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 597
Re: DXO OpticsPro 10 vs. ACR
« Reply #1 on: November 29, 2014, 03:45:55 pm »

Here is an attempt to make some sort of comparison.  All of the caveats apply.  I opened your DNG in ACR 8.7 (PS CC) and there were ACR settings embedded in the file - I assumed that those were your "optimal" ACR settings that produced a file that you found acceptable.  I then tried to emulate the resulting image (rendered to PS in ProPhoto) by beep boop bopping in DXO.

Attached is a screenshot of the two resulting images, rendered and opened in PS, 2-up vertical, fit to window.  The haloing that is apparent around the border between the tree and sky areas is amplified visually at this zoom level and is not so contrasty at 100% zoom.  The side-by-side is just to give you an impression of the comparison in overall tonality, etc.  See 100% res files (links in post below).

Some general comments.  The blue and green channels are blown in much of the central sky area.  This makes the resulting raw conversion highly dependent upon highlight reconstruction - i.e., not recovery with negative exposure, but reconstruction from secret sauce recipes for rebuilding areas of tone from unblown pixels in other channels.  In this case, the blown areas are, more or less, neutral, so it should be fairly straightforward from a luminance standpoint.  That said, DXO has never been good at highlight reconstruction, and ACR/LR perform compression and reconstruction automagically that the user cannot, to a certain extent, control.  So the end result is somewhat reliant on the user's ability to coax the data into some visually acceptable form, especially given the lack of data in the cloud area.  You can probably get something "acceptable" but it may not be very accurate.

I am not a DXO fanboy but I use it and other raw converters (RPP, Iridient Developer, dcraw, Raw Therapee, ACR, C1).  They all have their quirks, strengths and weaknesses.  That said, DXO has a pretty vast array of controls that make the learning curve steep.  Plus, the controls often interact, again making their impact not so obvious.  The two major adjustments made to your DNG in DXO were 1) disabling Smart Lighting and -2.3 stops exposure and 2) changing the Color Rendering to "neutral color, realistic tonality (gamma 2.2)."  These two adjustments flatten the contrast of the image and give you a better starting point for redistributing the tones and adding local contrast.

As far as the funkiness of the Chromatic Aberration and artifacts in the trees on the right - firstly, you should enable all of the options in the CA tab; secondly, turn down the Lens Softness "Details" control.  The same funkiness is present in the ACR image, but rendered slightly differently.

DXO, in general, is set up out of the box to make a lot of automatic corrections, even on a per image basis.  For a very challenging image like this, it is often better in DXO to start with the "No Correction" preset.  Sometimes it is easier to get where you want to go with an image to remove all of the corrections and go through your progression - WB, tone, contrast, color, etc. - manually.

In the post below are links to full-res JPEGs, converted to sRGB, of each conversion.  I used the normal, "High Quality" noise reduction (Lum NR = 12 and 100 chroma NR) on the DXO image but I did not sharpen the DXO image.  I did not make any perspective or optical corrections.

Overall, the ACR version appears to have slightly more perceivable detail in the sky compared with the DXO version.  If you are not familiar with DXO, the ACR result would also take you a gazillionth of the time the DXO result would take you to achieve.

One thing that is very apparent is that the ACR version suffers from splotchy green and magenta chroma artifacts in the shadow tones throughout the image and, overall, the detail is somewhat smeared in many of the mid and shadow tones int he ACR version compared to the DXO version.  I would say that, in my two conversions, the DXO image has better shadow detail and color.

This is a very challenging image to acquire and convert in pretty much any raw converter, with the 5D3.

kirk
« Last Edit: November 29, 2014, 04:00:40 pm by kirkt »
Logged

kirkt

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 597
Re: DXO OpticsPro 10 vs. ACR
« Reply #2 on: November 29, 2014, 03:49:30 pm »

Here is the ACR image link (Dropbox):

https://dl.dropboxusercontent.com/u/78890747/ACR.jpg

and the DXO image link:

https://dl.dropboxusercontent.com/u/78890747/DXO.jpg

They are both about 3.7MB images.

kirk
Logged

kirkt

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 597
Re: DXO OpticsPro 10 vs. ACR
« Reply #3 on: November 29, 2014, 03:54:05 pm »

Raw digger info.

kirk

Logged

BernardLanguillier

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13599
    • http://www.flickr.com/photos/bernardlanguillier/sets/
Re: DXO OpticsPro 10 vs. ACR
« Reply #4 on: November 29, 2014, 07:33:31 pm »

How about this quick conversion with C1 Pro 8?

https://www.dropbox.com/sh/uxhw240j11n277i/AACihMBezVZrH2lwB642Oipba?dl=0

It feels more natural to my eyes.

I used the highlight recovery on the base image and then created 3 layers:
- one to brighten up the trees in the upper right,
- one to brighten up the foreground,
- one to brighten up the tower.

Cheers,
Bernard

Redcrown

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 431
Re: DXO OpticsPro 10 vs. ACR
« Reply #5 on: November 30, 2014, 12:13:47 am »

kirkt,

Thank you very much for your reply and considerable effort. It's rare these days to find someone who is both willing and able to address the kind of issues raised here.

Your DXO version is a very impressive match to the ACR version, and allows for a better starting point to compare the two programs. I never thought to try a big negative exposure in DXO. But when I follow your steps I see where that gives a decent rendition of the sky.

However, I still can't come close to your DXO version. When I follow your steps and use a -2.3 exposure, no Smart lighting, and the "neutral color, realistic tonality (gamma 2.2)" rendition, I get a very dark image, of course. So I start fooling with shadows, blacks, contrast, etc. Even tried Clearview and a Tone Curve. So I remain a little confused. Would you mind sharing what other DXO adjustments you made to get to the jpeg you posted. Maybe even dropbox the "dop" file?

Looking at your two jpegs (at 200%) I agree that the shadow rendition in DXO is better than ACR. That's the first DXO virtue I've seen, so I'm encouraged to keep testing.

I've seen various claims that the ACR highlight recovery works by using good values in any unclipped channels to reconstruct values in clipped channels (secret sauce recipies). Based on my experience I tend to believe that. It's one of the reasons I picked this image for testing. If you look at the individual channels in RawDigger you see there is very little clipping the in red channel, much more in the blue channel, and even more in the green channels. Should be a good condition to see the difference between "reconstruction" and "recovery" of highlights.

Also (for others), I'm not trying to get a good rendition of this particular image. It's just one exposure from a bracketed set that produced good results via HDR. I'm just using it as a sample to test and learn DXO.
Logged

kirkt

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 597
Re: DXO OpticsPro 10 vs. ACR
« Reply #6 on: November 30, 2014, 03:43:11 pm »

Hi Redcrown - you are welcome.  This is an important exercise, regardless of whether you want to produce something "finished" or a neutral, flat starting point for further editing.  I figured the point was to see how much data you could extract from a single exposure.  I shoot with a 5D3 as well, after shooting with a 5D and then a 5D2 for many years.  I have recently been trying to understand just how much data one can squeeze out of a 5D3 image under a variety of shooting conditions.  

A couple of observations about DXO versus ACR and raw converters in general.  One of the things that I think is initially laudable is ACR's ability to reconstruct highlights from seemingly nothing.  Especially for folks trying to walk that fine line of maximizing ETTR, this extra "secret sauce" in the highlights, along with ACR's automagic highlight compression gives one the impression that ETTR can be pushed pretty far before noticeable loss of data.  (Just open your posted image in ACR and switch between PV 2010 and PV 2012.  How did all of those clipped highlights shown in PV2010 get not-clipped in PV2012?)

And that's OK if all you ever plan to do is use ACR/LR for your conversions.

But I would also argue that your raw converter might start breeding bad exposure habits if you rely upon this magic - should you need or decide to use another converter, this might cause all sorts of problems in the important highlights.  This uncertainty in the extreme highlights, coupled with ever increasing quality of noise reduction has made me re-examine my ETTR practices so that they are, to a certain extent, raw converter agnostic.  That is, I can recover all of the highlight data my ETTR captures by pulling the image in conversion.  I can then use the various tone mapping tools in raw conversion or in PS to redistribute tone, with less noise in the shadows that is afforded by ETTR.  A third or half stop extra ETTR will probably not affect shadows significantly but can destroy important highlight data if you push it a little too far.

As I started using DXO in v7, I actually started exposing for highlights and letting shadows fall where they may - eventually I found a middle ground that retains all of the highlights with ETTR, and relies on a little bit of noise reduction if necessary for extreme pushes in raw conversion.

The real breakthrough in all of this is Magic Lantern's raw histogram.  No need to guess any more about when the green channel (typically - or any channel) is clipping.  The raw histogram gives me the threshold for ETTR and then I typically back off 1/3 of a stop or so.  This way there is no need to rely on highlight reconstruction, which can vary from converter to converter in quality and accuracy.  That's not to say that I never ETTR with some clipping, maybe that is appropriate for the scene.

Bottom line, I'd rather recover highlight data that is already there than have the raw converter fabricate it.  If need be (say in the example you posted) I can extract the red channel of the linear TIFF version of the file and apply that to the green and blue channels in PS where necessary to get neutral highlight detail back.  Clipping in full color areas is a bigger problem, and that may be where highlight reconstruction with various techniques could be useful (Raw Therapee).  

Here are links to the DXO Preset and DOP files - they may not correspond exactly with what I ended up posting, I can't remember exactly what small tweaks I may have made after saving the preset.  it is definitely a lot easier to provide these files than describe what I did to tone map the image.  You will note on the Tone Curve that there is a "gamma" field that applies a gamma curve to the tone curve - another nice feature.

Preset:

https://dl.dropboxusercontent.com/u/78890747/CastleDNGPreset.preset

DOP:

https://dl.dropboxusercontent.com/u/78890747/ire038.dng.dop

kirk
« Last Edit: November 30, 2014, 03:52:31 pm by kirkt »
Logged

Redcrown

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 431
Re: DXO OpticsPro 10 vs. ACR
« Reply #7 on: November 30, 2014, 04:39:30 pm »

Kirkt,

Thanks again. I've loaded the preset/dop you posted and can see all the adjustments now, including the tone curve. I need to spend more time studying and playing with them, but they will give me better insight on how to work with DXO.

We have followed a similar path. I started digital with a Canon 1Ds, went to a 5D2 and then a 5D3. I've worked on a handfull of Nikon images from friends, but mostly I'm Canon guy and constantly working to get the most out of that gear.

You've also got me re-interested in Magic Lantern. I've not used it yet, but have browsed a lot of info and Youtube videos in the past. Most of what I saw was about shooting video, which I don't do, but now I will probably try it, just for the raw histogram feature.
Logged

kirkt

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 597
Re: DXO OpticsPro 10 vs. ACR
« Reply #8 on: November 30, 2014, 05:06:56 pm »

Re: Magic Lantern.  In addition to the ML raw histogram, there is a module that you can enable to shoot what is called "DualISO" - if dynamic range is super critical, and you do not mind sacrificing some resolution (most of the time not noticeable) you can set up your camera to shoot a single exposure and have the sensor read off at two alternating ISOs simultaneously.  Two lines are read at the primary ISO and then two at the recovery ISO.  You can set the primary ISO to be for highlights (say ISO100) and have the recovery ISO be for shadows (say ISO 1600) - or vice versa, set the primary for the shadow ISO and let the secondary ISO recover highlights.  This produces your primary highlights with less noise in the shadows (or vice versa).  The resulting CR2 file needs to be run through a utility called cr2hdr - it outputs a DNG that has the best pixels blended together.  It is an on-camera approach similar to the venerable Zero Noise application.

Give it a shot.  DXO and ACR can read the resulting DNGs, so you can use either to tone map the results.

kirk
Logged

robgo2

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 388
    • Robert Goldstein Photography
Re: DXO OpticsPro 10 vs. ACR
« Reply #9 on: December 03, 2014, 12:41:50 am »

Here is what Photo Ninja can do:  https://www.dropbox.com/s/x9zc4eg2uukibry/ire038_v2.jpg?dl=0

Rob
« Last Edit: December 03, 2014, 12:56:45 am by robgo2 »
Logged

Wayne Folta

  • Newbie
  • *
  • Offline Offline
  • Posts: 2
Re: DXO OpticsPro 10 vs. ACR
« Reply #10 on: December 14, 2014, 01:19:35 pm »

I tried my hand at RawTherapee, but couldn't quite match the color:

https://www.dropbox.com/s/q5kmenond4rcutk/ire038.jpg?dl=0

your sky is cooler, but your tower/vine is warmer. Then again, I'm a beginner at these tools. I like Darktable better than RawThrapee in several ways, but couldn't get close to your examples with it at all.
Logged
Pages: [1]   Go Up