Have just done a reasonable sized project with Face Detection in Feb 2023 - the first time since the December update. There is no mention of any changes in Face Detection in the release notes so it is unclear to me why the editing experience felt a little better - but there is still room for improvement. I could be imagining things too.
It seemed that if you had a sequence of images making corrections to the face labels may have seen improvement in adjacent images but only if you paused for a while and came back to that image. Specifically, I think some images had ? alongside. When those images where corrected with a name, adjecent images still with the ? might have updated after a short period.
There was a special guest that appeared in many of the photos. However, even if you made the corrections to an incorrect labelling, the results didn't seem to improve. It is seems that face detection is only leveraging the facial features and doesn't take into account clothing. The guest is the only one wearing a green jacket.
The UI for editing could have more white space - it is easy to mis-click the check mark and trigger the text edit view, where you can easily accidentally delete the correct text. Because you are focussed on one label, there is ample space to separate out the confirm / ignore buttons - they don't have to be compressed so tightly for this task.
On an Apple Magic Mouse, a scroll event can be immediately detected when you are confirming the face detection result with the tick box, so you end up scrolling left/right images when you don't want to. There's clearly no time delay between acknowledging the tick and responding to scroll events. You end up on the wrong image because LRC has just scrolled a lot of images.
I felt like I needed to know that if I made a change that something was being recalcuated for the next few images. There were a couple of faces that consistently were wrong. Even though you made the corrections, the shots taken a second or two later were never corrected. So updating the recognition estimates for temporally close images didn't seem to be happening.
If you accidentally make a typo and there are now multiple keywords for the same person, the process of going back and finding those images is non-intuitive since you have to search via the Keyword list. This gives you a filtered view of just the selected images. Once you make the correction, you are left with an empty view with no obvious way back to where you were - there's no back button.
12.1 LRC macOS Monterey
AdChoices