Participant

Question

Photoshop Element Organizer label grayed out people recognition

Forum|Forum|9 years ago
April 22, 2017
3 replies
2730 views

In Photoshop Element Organizer v15 I launched the recognition of faces.

Overall it works well.

But some labels appear grayed out unless I understand why.

Several cases:

-In 'All people' the face is displayed and the name is written below.

Yet when I click on the label, in the section "Faces confirmed", some labels are grayed out.

This is surprising since these face are confirmed (recognized)

and that by landing on a gray label the corresponding photo (with the right face).

-In 'All people' label is gray with the name written below.

When I click on the gray label, all labels "Faces confirmed" are grayed out.

And when I ask myself on a gray label, the corresponding photo appears (with the right face).

= What is the problem?

- And how to fix it so that all the labels display correctly?

Thanks for your help

Alain

This topic has been closed for replies.

H

Hiew5E75

Participant

Browse

Greg_S.

Community Expert

By "other features," I mean using the UnNamed People room to add names to stacks of recognized faces, using Don't Show Again feature to ignore faces that are not going to be named, and Confirming faces in the Named People room under the "Is this X" dialog. This, for me, is the most efficient way to tag People. I occasionally also add names in the Media grid.

As for deleting photos and importing them again to improve FR, I don't think that would be of any use. Face recognition is just that - recognizing a face within a photo and identifying its size and location within a photo. Adding a name to a face does not change the program's ability to recognize whether or not there is a face in the photo. Nor does deleting and reimporting a photo after another recognized face of the same person has been identified improve the process. As I understand it, once you add a name to a stack of photos, if you open the Add Name text box of another stack of the same person, a name/names will be auto-suggested for that person. Also, in this situation, if you go to the Named People room, there may be a blue triangle next to the identified person's stack, and if you open that named person's stack, you will see suggestions of other photos of that person ready for confirmation. Those "is this X" confirmations are for photos that the program now thinks is the same person based on your naming that person. If you confirm the face here, the same photo of that person will be removed from a stack in the UnNamed People room.

The only reason I can think of for deleting and reimporting a set of photos is if you have mistakenly hit the Don't Show Again button/command. In that case, the face will be removed from consideration as a recognized face and would need to be imported again for the face to be recognized. (Or the Mark Face tool can be used to add the face manually.)

N

Neb1

Known Participant

Thanks for your fast response.

Greg_S. said:

By "other features," I mean [...]

Good to hear, as this means that I am already using all the features that FR offers and that help FR in automatically recognizing as many pictures as possible.

Greg_S. said:

I occasionally also add names in the Media grid.

OK, this might explain our different perception of how much the user has to spend for marking faces. I understood from earlier posts in this thread that adding names in the Media grid is not the ideal solutions as this leads to grayed out icons and does not improve the algorithm's skills (via self-learning) so I understood the only "true" option is Mark Face, which takes orders of magnitude longer than plain taggingof (multiple) pictures in the media grid.

Greg_S. said:

Face recognition is just that - recognizing a face within a photo and identifying its size and location within a photo. Adding a name to a face does not change the program's ability to recognize whether or not there is a face in the photo.

Just to avoid misunderstandings: In my step 2 with "manual marking of faces not recognized" I was not referring to tagging in the media grid but to Marking Faces. In this process you tell the program the "size and location" of a face within a picture just as you wrote and not just simply adding a name to picture. Given your earlier sentence "Yes, the algorithm is self-learning. The more faces you confirm of a single person, the more FR recognizes that person." it is not obvious to me why you think that the 5 steps given above should not lead to better FR results, as you could regard the Mark Face thing also as a "confirmation" in your words that person X is in this picture which should help the FR algorithm, shouldn't it?

Greg_S.

Community Expert

I understood from earlier posts in this thread that adding names in the Media grid is not the ideal solutions as this leads to grayed out icons and does not improve the algorithm's skills (via self-learning) so I understood the only "true" option is Mark Face, which takes orders of magnitude longer than plain taggingof (multiple) pictures in the media grid.

There is also the option to hover over a recognized face in the grid and add a name there.

you could regard the Mark Face thing also as a "confirmation" in your words that person X is in this picture which should help the FR algorithm, shouldn't it?

It seems to me that you only need to use Mark Face when the program can't even recognize that a face exists in the photo. This can be for a variety of reasons, including lighting and shadows, the angle of the face/body, glasses or goggles, hats and scarves - and in this day and age: face masks. So adding a name to something that the program only sees as an unidentified blob does not necessarily mean that the algorithm will identify another blob as a face and certainly not a face belonging to X. Perhaps AI does a better job than I think. But I can't imagine that it is relying on users repeatedly deleting and importing the same set of photos.

A

Anwesha Dev Chaudhary

Participating Frequently

Can you please attach a screenshot highlighting the issue?

Are you working in Named view or Unnamed view?

Thanks,

Anwesha

A

alainclnAuthor

Participant

Below, screenshot highlighting the issue.

I'm working in Named View

Greg_S.

Community Expert

Dear Greg,

thank you very much for your help. I followed your instructions (mark face button, bounding box etc.) for all the photos which I dragged and dropped before, and when I clicked on the ppl tab, I saw the generic icons for a sec or so and then they got automatically replaced by the face I manually marked. Problem solved!

(In a way, it makes sense that generic icons appear as they appeared because if Organizer does not recognize the face, it cannot know what to display there - maybe the complete photo would be a better idea though rather than a generic icon.)

Anyway, the drawback of your solution is: It takes ages to do it manually (I guess this is no news for you), and there are many pictures I'd have to do manually. Hence, my questions I did not find answers to it so far:

1. Is the 2020 face recognition algorithm self-learning, so getting better with those people I am helping it with?

2. Is there any Adobe application that has a better algorithm or is it all the same acrosse their apps? I understand, that the cloud services use Adobe Sensei, and on the Elements marketing site it says FR is "based on", "supported by", "powered by" Sensei. Is FR within Elements the same as Sensei or is it worse?

3. How can the FR progress be seen? I found in another post of yours the hint to count the number of json files in the face analysis folder in the catalog; this is 53 items less than media items in the catalog (there are more than 53 videos in the catalog, so this cannot be the cause for the difference). If one face in a picture is recognized, but the second is not, does it necessarily mean that I have to do the second manually because the algorithm is finished with the picture or is this consequence not always true?

Yes, the algorithm is self-learning. The more faces you confirm of a single person, the more FR recognizes that person. But, as I say, it is not perfect, and it has difficulty picking up which of my two grandsons is which, particularly when they were younger.

I don't know whether Lightroom uses a different algorithm or not. I didn't like the FR interface so have not used it much. As I understand it, Elements uses "Sensei", but I believe that the FR technology is actually licensed from a third party.

Elements 2020 analyzes photos first before turning to analysis of video. The following status will appear when the analysis is ongoing:

You haven't mentioned whether you were aware of the Hide Small Stacks checkbox. Did that exose more photos for you in the UnNamed People tab?

As for the speed of tagging, I have been using the program for many years and I became very adept at quickly selecting and tagging photos. In the early days of FR (which started in Elements 11 I believe), I could tag my photos faster manually. But, now, the program is so efficient that it is much faster for me to use the UnNamed People room to tag faces even though there may be many thousands of stacks of irrelevant faces.

Sign up

To post, reply, or follow discussions, please sign in with your Adobe ID.

Sign in to Adobe Community

To post, reply, or follow discussions, please sign in with your Adobe ID.

Scanning file for viruses.

This file cannot be downloaded