Lightroom/Camera Raw: Auto Tone Applied Incorrectly to Cropped Images

  • 4
  • Problem
  • Updated 1 month ago
  • Acknowledged
  • (Edited)
Prior to Lightroom 7.1 cropped image Auto Tone settings were calculated the same as uncropped images using the full image data. Lightroom 7.1 introduced Adobe Sensei technology and calculation of Auto Settings using only the crop image data. By excluding the area outside of the crop it should improve the image rendering since it uses ONLY the target image data inside the crop for calculating and applying the Auto settings.

A number of people are still complaining that the new and improved Auto Settings is not working well with certain image types. The most common complaint is that it still under or overexposes the image. While working on another similar problem I discovered an issue with the way Auto settings are calculated and applied to cropped image files. It appears the Auto Settings values are being calculated properly using only the crop area image data, but then applied "adaptively" using the entire uncropped image data as the basis for applying the new image data values.

This causes cropped images with much darker or lighter image content outside the crop area to render darker and lighter respectively. Obviously, this makes the Auto Settings function less accurate than it could be using only the crop area image data for all Auto Tone processing steps.Below is a controlled test I ran to confirm this behavior. I used a large piece of poster board with gray on one side and white on the other to simulate lighter and darker subjects outside the crop area. Manual exposure was used to insure the two image files (gray board and white board) have identical camera exposure values (shutter speed and aperture). What I noticed is that underexposed image file pairs exhibit larger differences. The controlled test files used here are ~1.75 EV underexposed, but the normal exposure file pairs shot at the same time also exhibit the issue. You can download the two CR2 files used in the below screenshots here.

https://www.dropbox.com/sh/44gpzg9skb1g7lt/AAC-5i5UBSQUrlfhXsXDvwJVa?dl=0
Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes

Posted 1 month ago

  • 4
Photo of Yves Crausaz

Yves Crausaz

  • 425 Posts
  • 60 Reply Likes
Same for me under macOS,
Je viens de faire les essais sur 2 copies virtuelles du nême RAW original, mais une des copie est recadrée sur le sujet principal.
Les valeurs définies par la fonction Auto sont différentes pour les deux copies virtuelles. Mais est-ce gênant ? Je n'en suis pas certain, peut-être faut-il développer différemment si recadrage ?

I have just done the tests on 2 virtual copies of the original RAW, but one copy is cropped on the main subject.
The values ​​defined by the Auto function are different for the two virtual copies . But is it annoying? I'm not sure, maybe it should develop differently if cropping?


Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes
That is correct behavior and what was added in LR 7.1. Original full-size image files and cropped virtual copies will render differently and more accurately with the new Auto Settings, which is why it was introduced. The problem I'm reporting here is that the cropped image Auto Settings function uses the cropped image data for the settings calculation, but then applies those settings as if the image was NOT cropped. It's a two-step process that requires using the cropped image data for BOTH steps. This is due to the "image adaptive behavior" of the Process 3, 4, and 5 Tone controls.

To test this for yourself you can download the controlled-test CR2 files at the below link.

https://www.dropbox.com/sh/44gpzg9skb1g7lt/AAC-5i5UBSQUrlfhXsXDvwJVa?dl=0
(Edited)
Photo of Yves Crausaz

Yves Crausaz

  • 425 Posts
  • 60 Reply Likes
I do not understand your reasoning, if you reframe, or you want a new analysis, or you keep the analysis done before the reframing? Personally, I create a virtual copy and crop in this virtual copy and I do a new analysis, so my two crops have different values, it seems to be a correct operation, obviously only if this cropping is important!


Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes
Personally, I create a virtual copy and crop in this virtual copy and I do a new analysis, so my two crops have different values, it seems to be a correct operation, obviously only if this cropping is important!
Yes that is correct operation and you would expect the cropped image to have different Auto Settings values than the full-size original image. The problem is that the cropped image Auto Settings values are calculated using the cropped image data, but then applied using the full-image data. I know it's difficult to grasp exactly what's happening until you fully understand the problem.

To demonstrate the issue you need to use two image files shot of the same central subject that have different toned surround area. Please download and test the two image files I have already created that easily demonstrate the issue.

https://www.dropbox.com/sh/44gpzg9skb1g7lt/AAC-5i5UBSQUrlfhXsXDvwJVa?dl=0

1) Measure the grayscale patches in the two ColorChecker CR2 images files, which should be virtually identical since they were shot with the same manual exposure settings under the same lighting condition.

2) Next create virtual copies and crop both image files to show only the ColorChecker as in my screenshots.
3) Apply the Auto button in the Basic panel to the cropped virtual copies. The images should appear identical and the grayscale patches have matching values, but the don't!

That is the problem being reported here. Does that make sense now?
(Edited)
Photo of Yves Crausaz

Yves Crausaz

  • 425 Posts
  • 60 Reply Likes
I tried with a crop on the dark part of the image, indeed AUTO gives me an over-exposed image compared to the original, but hey, I told myself that when a picture is very dark, AUTO made in general the image too bright for my taste.

Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes
You did not follow the three step instructions I provided above that demonstrates the issue.

Vous n'avez pas suivi les instructions en trois étapes que j'ai fournies ci-dessus et qui illustrent le problème.

Pour illustrer le problème, vous devez utiliser deux fichiers image du même sujet central ayant une zone surround différente. Téléchargez et testez les deux fichiers image que j'ai déjà créés et qui illustrent facilement le problème. Vous n'avez pas suivi les instructions en trois étapes ci-dessus qui illustrent le problème.

https://www.dropbox.com/sh/44gpzg9skb1g7lt/AAC-5i5UBSQUrlfhXsXDvwJVa?dl=0

1) Mesurez les niveaux de gris dans les deux fichiers d’images ColorChecker CR2, qui doivent être pratiquement identiques car ils ont été pris avec les mêmes paramètres d’exposition manuelle dans les mêmes conditions d’éclairage.

2) Créez ensuite des copies virtuelles et rognez les deux fichiers image pour afficher uniquement le ColorChecker comme dans mes captures d'écran.
3) Appliquez le bouton Auto du panneau Basique aux copies virtuelles recadrées. Les images doivent sembler identiques et les correctifs en niveaux de gris ont des valeurs identiques, mais ce n’est pas le cas!

C'est le problème signalé ici. Est-ce que cela a du sens maintenant?
Photo of Yves Crausaz

Yves Crausaz

  • 425 Posts
  • 60 Reply Likes
I see, but is Adobe aware of that? And what to do now? Are you asking for a change of function if it is possible?

Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes
Rikk Flohr is an Adobe employee (see below) so yes Adobe are now aware of the issue.
Are you asking for a change of function if it is possible?
It's my understanding that the new Auto Settings function introduced LR 7.1 uses Adobe Sensei technology and calculation of Auto Settings using only the crop image data. Clearly that is not the case from my testing so I am requesting it to be fixed.

https://helpx.adobe.com/lightroom-classic/help/whats-new/2018.html#LightroomClassicCCv71December2017

Adobe's Auto feature now uses advanced machine learning to automatically apply the best edits for these slider controls in your photos: Exposure, Contrast, Highlights, Shadows, Whites, Blacks, Saturation, and Vibrance. In addition, the Auto feature now also includes the ability to optimize the adjustments of the photo even after cropping has been applied.


(Edited)
Photo of Rikk Flohr

Rikk Flohr, Official Rep

  • 5600 Posts
  • 1184 Reply Likes
@Todd, does Adobe Camera Raw exhibit the same behavior? Trying to figure out which team gets this...
Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes
Rikk, I see the exact same behavior and readings using Camera Raw 11.3 with the two files posted to Dropbox.
Photo of Rikk Flohr

Rikk Flohr, Official Rep

  • 5600 Posts
  • 1184 Reply Likes
Generally speaking, any time there is weirdness in Develop, I check it against ACR. If Camera Raw has the same issue the bug gets filed against Camera Raw.  If one of the Camera Raw clients has the issue, we file it against the client.  I have pinged the Camera Raw team for comment.
Photo of Todd Shaner

Todd Shaner, Champion

  • 1464 Posts
  • 492 Reply Likes
Thanks Rikk!
Photo of Roelof Moorlag

Roelof Moorlag

  • 189 Posts
  • 57 Reply Likes
Good work Todd!