• Global community
    • Language:
      • Deutsch
      • English
      • Español
      • Français
      • Português
  • 日本語コミュニティ
    Dedicated community for Japanese speakers
  • 한국 커뮤니티
    Dedicated community for Korean speakers
Exit
0

P: Increase width of Text Pop up for Suggested Person - Face Detection.

Contributor ,
Sep 27, 2022 Sep 27, 2022

Copy link to clipboard

Copied

Some ideas to improve the Face Detection workflow.

 

When manually updating the name of the suggested person, the text pop-up is too narrow. It could be double the width to minimise truncation of the whole name. The width should probably be based on some localisation settings or be content driven i.e. match the widest length.

 

Additionally, the text pop up could be enhanced with at least one thumbnail to help me identify the person. It could be optimised to help the photographer choose more powerfully with either text or image.

 

When in the Faces view and you are asked to confirm faces, it would be good to understand context and actual size of the person in the photo. People in the background who are slightly blurry are possible options. However, you want to make a decision about whether they are validly defined in the photo. It's hard to do from uniform rectangles, noise from low light and pixels from being far away aren't easily distiguished.

 

There are two scenarios 1) you are screening photos and have a do not publish list so you need to filter out people, even if they are in the background, 2) you are trying to assess whether the face is actually the subject of the image rather than a bystander in the crowd. They are quite different scenarios with opposing needs.

 

I've mentioned this one before but Faces don't need to span the entire catalog. In many cases, the photos for a particular photo session have a finite closed set of faces, having the universal set makes the process inefficient. I might want to borrow some Faces from another folder but I should be able to choose to do that. A user scenario of event photographer should be considered - there are named people in my collection that I photographed 3 years ago that I will most likely never photograph in the next 5-20 years.

 

If you manually define a face rectangle, the system doesn't take an educated guess immediately (even on an M1). You have to manually type. There's an opportunity to deliver a quick visual cheat sheet based on the faces in the current collection (the first step is enhancing the pop up). I don't know the hundred or so people I photograph at an event, the interface needs to help me appear to be totally familiar 😀 - that's what UX design excellence is all about.

 

Idea No status
TOPICS
macOS , Windows

Views

111

Translate

Translate

Report

Report
Community guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
community guidelines
5 Comments
Contributor ,
Sep 27, 2022 Sep 27, 2022

Copy link to clipboard

Copied

Adding to this as I work through another 80 images. There's a weird interaction when you are correcting the name of a mis-recognised face. You might be editing and for some reason it doesn't move into edit state. So it is then sending keyboard events to other places. There's also another conflict if you are working quickly and are using the arrow keys to move between successive images.

 

There's a small subset of images that when completed revert to ? even though you just typed in and confirmed the person. So something subtle in the events. 

 

Overlapping text label areas are a pain. There needs to be a way to focus on just one. Also need to have them move out of the way whilst you look at the face. If there is a group, the labels are non-intelligently overlapping - plenty of scope here to think about non-overlapping geometries and using lines to connect face regions with text areas. You aren't trying to label everybody in a group at the same time. You are wanting to work on the unknown ones ... which could be a slightly different colour. (the new ruby lith / red masking UI and how it gets out of the way when you make adjustments is kiind of the focussed thinking that needs to be applied here).

 

 

Confirmation and rejection buttons could grow when hovered over. Small target = longer seek time.

 

Some images have the same person identified multiple times. Can't happen, without mirrors (or twins - now there's another flag). However, if you confirm one of the images, the others are likely in need of change.

 

From a workflow point of view, which images need my attention to fix? There's nothing at a higher level saying that this, you have to inspect every image.

 

Not taking advantage of the time delay between successive images. I have to essentially start from scratch even though the previous image was taken 3 seconds prior and has a lot of image correlation. There's a sequence of group shots with 20 people where I've also taken the same group from an oblique angle so it hasn't been able to pick out any faces.  - I want to take the previous face regions taken square on and paste them into this shot and manually adjust.

 

 

 

Votes

Translate

Translate

Report

Report
Contributor ,
Sep 27, 2022 Sep 27, 2022

Copy link to clipboard

Copied

Here's a repeatable bug

 

  1. Go to click on a tick mark to confirm but because of the proximity you click just inside the text field, activating edit mode (so the selection highlight is around this text)
  2. Think that you've already confirmed the previous one and go to confirm another face tick box

 

When you do this the previously selected text is assumed to be incorrect and it reverts to ?

 

 

So you lose the correction on the first face and it reverts to ?

 

The problem is that there isn't enough space between the confirmation buttons and the text box.

 

(Mac)

 

Votes

Translate

Translate

Report

Report
Contributor ,
Sep 27, 2022 Sep 27, 2022

Copy link to clipboard

Copied

If you have first name and last name, there is no way to partially type the last name to search in the face text field. It only works with starts with. "G Washing" doesn't produce a search result. A smarter search which could have partial first name and partial last would help or just some initials to speed things up. I'm assoicating faces with last names because they are less common.

Votes

Translate

Translate

Report

Report
Contributor ,
Sep 27, 2022 Sep 27, 2022

Copy link to clipboard

Copied

When backspacing / deleting an incorrect face label, the pop up never re-appears to give you possible matching selection. You have to type something for the options to appear again. At the very least, a timer could be triggered to detect when you have finished backspacing and then reveal the pop up menu again. It's clumsy.

 

There's also a weird time delay that if you are backspacing it may go before the string and then ask you whether you are deleting the image. There has to be a saftey net to prevent falling instantly into a deletion state.

 

On deletion, if you have a crowd shot with lots of faces that you don't need, they have to be deleted one at a time. The time delay to delete is super long in terms user interface time frames. It isn't responsive.

 

Up to about 400 images confirming faces ... another 150 to go....

Votes

Translate

Translate

Report

Report
Contributor ,
Sep 27, 2022 Sep 27, 2022

Copy link to clipboard

Copied

LATEST

If you are keyboard focussed and working quickly, the text field accepting the name update can miss the first few characters when you are typing. You've already clicked the text box to activate but it isn't fast enough to pick up the characters that you have started to type.

 

There's also a timing delay with the pop up menu that appears, it can appear but then clear what you have already partially typed.

Votes

Translate

Translate

Report

Report