Masking bugs in ver 13.0.1 - Mac Mini M2Pro, macOS Venture 13.5.2

[MightyAtomBoy]

Potential AI bug encountered when pasting masks to multiple images. Linear gradient not detecting same area despite identical composition (tripod shot, same frame/environment, slight move in subject). Incomplete masks show a solid object with masking through only half. Edge detection failing?

Steps:

1. Apply multiple masks to image - eg, linear gradient, radial gradient, etc

2. Select multiple images > synchronise settings

[MightyAtomBoy]

In addition: Mask settings not pasting entirely to some images.

[johnrellis]

To see if this is a general problem in LR or something specific to your configuration, select a sample source photo with masks and a sample target photo, do File > Export As Catalog with the options Export Selected Photos Only and Export Negative Files. Zip the exported catalog folder, upload it to Dropbox, Google Drive, or similar, and post the sharing link here.  We can see if this issue happens on other installations, and if so, get an actionable bug report filed.

[MightyAtomBoy]

Hi John. Thanks for the assist. Link here to the files requested: https://drive.google.com/file/d/1nWJVQHlEkrW0aRvY7Mni3zbiFXXnKOdu/view?usp=sharing

I'm also encountering other issues beyond masking when synchronising settings, such as WB not pasting.

[johnrellis]

[This post contains formatting and embedded images that don't appear in email. View the post in your Web browser.]

Very sorry for the delay in getting back. I misplaced the email notification.

I changed the overlay color from red to green to make it easier to see what's going on with the seats.

I looked at the individual mask components (Linear Gradient, Subject, Radial Gradient), and the Linear and Radial Gradients look fine. The issue is the funky Subject mask on IMG_4049, which has picked up parts of the seats, but on IMG_4050, it's fine:

There are many posts here observing that Subject masks often make mistakes with people. Usually you get better results using People masks, which look like this on the two photos:

It didn't pick up the dark knee in IMG_4049, but otherwise they're identical.

When I replace the Subject mask in IMG_4049 with a People mask and then copy all the settings to IMG_4050, I get good results. Similarly if I reset IMG_4049 and copy from IMG_4050.

[MightyAtomBoy]

Hi John,

Thanks for the comprehensive reply. An intersting outcome. Can we assume the subject mask is less rigorous, or calibrated differently, in its calculations? If it's not detecting a straight line intersecting an organic shape (the seat back with the subject, clothing, etc), I figure the People mask is calibrated for the irregularities of the human form - as you suggest. I'll do some comparison tests on the rest of the images (we shot a few hundred over eight differenent setup and angles in the same space) and modify my workflow.

Thanks again for your help!

[johnrellis]

The Subject model is trained to recognize a very broad spectrum of what humans would consider the primary "subject" of a photo: a human, a flower, a motorcycle, a barn, etc. Whereas the People model is trained just on humans. So the People model trades breadth in what it recognizes with accuracy of what it does recognize.

[MightyAtomBoy]

I see, understood. The definition of 'subject' for me veers more on the cinematic 'subject', the person. Now I know!