Forget a face?

[George Koulomzin]

I would like to add some concrete suggestions right off the bat:

1.Issues in the People view:

a.The scope of face detection/recognition (i.e. the set of faces being 
processed) is either not working correctly, or the semantics are 
completely unclear, i.e. I am unable to restrict either detection or 
recognition to a specified set.

b.It is confusing to use the same set of tabs to define detection and 
recognition behavior.  The semantics are different. The search space for 
detection is sets of images.The search space for recognition is sets of 
detected faces. Re-scanning for face detection affects face recognition, 
but re-scanning for recognition does not affect detection.Handling of 
conflicts between one scan and another differs.

c.The "scan collection for faces" button sounds like it is detecting 
faces, but is used in both cases.. What "collection" is it scanning 
(entire collection?  Selected face thumbnails? Selection criteria 
specified in tags? Tags specified in search bar?Some combination of these?)

d.What is the difference between the search bar (between face thumbnails 
and the three tabs) and the "search in"->"tags" option”The former 
selects only face tags, the latter can select any tag, including face 
tags.How do these specs interact?

e.The scope of detection and recognition should be independently settable.

f.The semantics of "search in" tab are unclear:

i.“Albums” operates on collections of images.

ii.“Tags” may operate on face tags or image tags (does the search work 
differently when face vs image tags are selected?

iii.The “search bar” seems to operate on face tags only.

iv.How do these criteria interact?

2.The term "scan faces" is ambiguous.  It is used in several places 
where it is not clear if it refers to detection or recognition.Also in 
the documentation.

3.It should be possible to set the face detection/recognition scope, 
workflow, and settings in clearly distinguishable ways:

a.the search criteria specified by the left side of the people view

b.the items selected in the right hand side thumbnail view, which allows 
manual selection

c.The item menu (which should be limited to face detection, or present 
all the possibilities available in the people view Workflow and Settings 
tabs)

d.People view Thumbnail context menu (which should allow only 
recognition related operations)

e.Other thumbnail context menus (which should work like c. above)

f.It should always be clear which method is in use. It might be fair to 
set some defaults in “settings”.

4.There should two classes of recognized faces.

a.those that are "normative", i.e. they contribute to the recognition 
algorithm (the model learns from them)

b.those that are "informative", i.e. identifications which are of value 
to the user, but might confuse the model (sunglasses, low quality images 
or view points).

5.There should be a mechanism to recognize/identify the same person's 
face at different stages of life, notably infant, child, adult, post 
accident or surgery, etc... (unless the model is powerful enough to 
recognize all as the same person - something the documentation suggest 
it cannot to.)It might be necessary to manually link face recognition 
models to the same name.

6.There should be a mechanism to delete face detection models

a.On a one-by-one basis through via context menu or Item menu.

b.On a set of models by any of the search

-- 
George Koulomzin
Hook Mountain Software Development, Inc.
7 Bridge Hill Lane,
PO Box 781
Bridgehampton, NY, 11932
(m) 914-393-6179
george at HookMountain.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/digikam-users/attachments/20231215/af3dc23f/attachment-0001.htm>