Option To Batch Apply Auto Settings With Full-Resolution Image Data

  • 7
  • Idea
  • Updated 1 month ago
  • (Edited)

At the below link it was discovered when batch applying the Auto Settings function to multiple image files the Basic panel control settings are sometimes different than when applying it to individual image files.

https://feedback.photoshop.com/photoshop_family/topics/auto-tone-of-a-batch-of-photos-gives-different-results-than-if-you-do-them-one-at-a-time

Lightroom 8.3.1 Process Version 5 Auto Settings has improved the image rendering considerably using its sensei Ai technology. For most image files the only manual adjustment required is a slight change to the Exposure setting. This is a huge timesaver when trying to get a large number of picture files edited for any purpose. Unfortunately, when using batch apply of Auto Settings Adobe Engineering forces the routine to use the lower-resolution camera raw cache image data. This is done because there is a large reduction in performance when using the full image data. There are often a number of image files that do not render well using the lower-resolution camera raw cache data, but are more than acceptable when Auto Settings is reapplied individually. YMMV dependent on the many variables of image file content, but the Auto Settings function in Lightroom 8.3.1 is pretty darn good when it uses the full image data.

Many people who process a large number of image files will start an Import process and do something else until it is completed. Users who have the time or run the import overnight will be better served if they can select a Preference option or other setting that allows using the full image data for calculating Auto Settings during an Import process or even for images already imported into Lightroom.

If there’s an issue with Imports needing the Library previews and camera raw cache created before Auto Settings can be run it could be queued to apply it after the Import operation has completed. A second Auto Settings option can be provided to rerun Library Preview building after the Auto Settings operation using the Import module’s setting for ‘Build Previews’ selection. This way the user can run the operation off-line of normal work or overnight and not have to invoke Preview Building manually after the Import operation.



Photo of Todd Shaner

Todd Shaner, Champion

  • 1458 Posts
  • 487 Reply Likes

Posted 1 month ago

  • 7
Photo of John R. Ellis

John R. Ellis, Champion

  • 4469 Posts
  • 1188 Reply Likes
I fully agree that users should have the option to choose quality over speed for batch Auto. I've compared Develop > Auto with batch Quick Develop > Auto on a large sample of photos, and there are significant, visually noticeable differences with at least 6% of photos (1 of 17).   

Full-quality Auto takes on average from 1.5 to 2.7 seconds per photo, depending on the photo's size. Many people would gladly pay an extra 2 seconds per imported photo to get noticeably better quality.

Adobe recently added Enhance Details that's computationally expensive and provides little or no benefit except in limited cases. In contrast, an option for full-quality batch Auto would be trivial to implement and would provide noticeable across-the-board improvement to everyone using Auto.

Some examples of the differences between Develop and Quick Develop Auto:





Comparing Develop > Auto with Batch Quick Develop > Auto

To quantify how much better Develop > Auto is than batch Quick Develop > Auto, I created a test set of 325 raw photos, with about 50 of the photos my own and the rest from the sample galleries for recent cameras at DPReview.com. The photos were taken by a number of photographers using 21 cameras in a wide range of conditions.

I imported the test set and waited for standard-sized previews to complete. Then I applied Quick Develop > Auto to the test set and exported the photos' Basic settings to a spreadsheet.  Next, I opened each photo in Develop and clicked the Auto button, and again exported the photos' settings to a spreadsheet. 

Here's a summary of the differences between Develop and Quick Develop Auto for each Basic setting, using various metrics:



For example, the maximum difference (absolute value) in Exposure between Develop and Quick Develop Auto is 0.60, and the 98th percentile has a difference of 0.20. That is, 2% of the photos have Exposure values that differ by 0.20 or more.

In the spreadsheet, I highlighted all the photo settings that were in the 98th percentile and then examined the photos with very large differences or that had multiple settings highlighted. I found 19 pairs of photos with quite noticeable visual appearances. There are probably more such photos in the test set, but I didn't examine all of them.  19 / 325 = 5.8%.

You can download those photo pairs here: 
https://www.dropbox.com/s/8s8a4yxq58zeera/autodiffs-2019-06-08.zip?dl=0

Each pair is named photo (the Develop > Auto version) and photo-qd (the Quick Develop > Auto version). Import them into LR and judge the differences for yourself.

That .zip also contains the spreadsheet "autodiffs.xlsx" containing the settings for each photo and summary analysis.

Photo of Todd Shaner

Todd Shaner, Champion

  • 1458 Posts
  • 487 Reply Likes
Thank you John for running this comparison test. Seeing these differences it begs the question as to how Adobe Engineering is creating the sensei Ai technology algorithms used for evaluating the image data. I would hope they are using the full-image data to develop and create those algorithms. As a design engineer with 50+ years of experience I take nothing for granted and propose the following suggestion.

If not doing so already I suggest Adobe Engineering investigate creating two separate algorithm sets for calculating Auto Settings using 1) Full-resolution image data and 2) lower resolution Camera Raw Cache and Library Preview image data. It may be possible to improve the faster lower resolution Auto Settings image rendering. Regardless, the full-resolution Adobe sensei Ai technology algorithms should be developed and created using the full-resolution image data to insure the best possible Auto Settings results.
Photo of John R. Ellis

John R. Ellis, Champion

  • 4469 Posts
  • 1188 Reply Likes
Some more examples, some of which weren't included in the downloadable .zip: