On 7/27/07, Stephen Kiel <[EMAIL PROTECTED]> wrote:
> Thanks for taking the time to reply to my request. It appears that you and I
> have a fundamentally different point of view on how to best select regions in
> an image. Let me throw out a couple of observations before I address some of
> your points in the hope that I can avoid starting a religious argument about
> which technique is better (especially if my side is outnumbered).
> - In most applications most of the users focus on about 20% of the options
> and capabilities in a given tool whether it is an office tool, or an
> engineering Design Automation tool. I suspect this is true of almost any
> tool that is fairly flexible. This does not mean that the 80% that any one
> user doesn't normally use aren't valuable as different users solving
> different problems.
> - It is important that the work on an image has an intuitive feel. No one
> methodology or selection method works best across all of the images you might
> encounter. Users will develop favorite methods based on their own success
> and failures with features of a tool. Failure with a feature of a tool is
> not a reflection on the feature, it just points out that a variety of robust
> features are necessary for a well rounded tool.
> - The "enhancement" that I am asking for is for the current mode of operation
> to perform in a manner that would be expected by a reasonable user. When the
> user selects a mode of operation using the menu and switches, it seems
> reasonable for the tool to honor those choices rather than making a
> unilateral decision to change modes based on cursor location. The fact the
> behavior is called out in the specification doesn't really change the fact
> that it is kind of user unfriendly.
The 'enhancement' that you describe, in my experience, has been
included in the GIMP for at least a year. I cannot reproduce the
problem you describe (and for the purposes of getting this changed,
you will probably need to come up with some reliable way of
> Let me try and address your points:
> >The Bezier tool is better suited to the tasks you're describing. Using
> >freehand tool as a precision tool (i.e. for background extraction) is a
> >bad idea.
> If I believed this I would not have bothered to ask for the enhancement in
> the first place. I think of using the freehand select for precision work as
> part of my methodology rather than a bad idea. When I started using Gimp I
> tried the Path / Bezier tool but in practice I really haven't found much use
> for it. The notion that it is "better" is subjective and is not in line with
> my experience.
> >Freehand tool is intended to make coarse selections or tweaks in
> >selections that don't need too much precision.
> This may have been the vision when the tool was put together, if it was I
> would say the Freehand select exceeded the original expectation. I believe
> that it is the best selection mode in Gimp for making precise selections.
> - How fine your control is for defining a selection line is determined by the
> level of zoom you are at, not whether you are drawing the line or defining it
> with a series of dots. The freehand select is as precise as the picture and
> tool will allow.
That doesn't even make sense. The freehand tool and path tool have
both exactly the same level of precision -- they both render polygons,
so they're infinitely-precise; freehand can be faster if you have a
steady hand; paths is more precise because it can draw shapes with
holes in them in the same step as drawing the outside - freehand
select must do such things in two steps, because it makes some
assumptions about how the user wants to use it.
Freehand select works under the exact same rules as path tool -
including eg. what happens when a polygon self-intersects. The only
difference is that freehand composes the path itself, so that you can
select a complex shape with click+move... instead of one click per
point as in paths selection.
To demonstrate that, move your mouse in a circle with freehand select,
and scribble through the circle before releasing the mouse button.
> - Describing a line with a series of dots is not inherently quicker or more
> precise than just drawing the line. If you have to start inserting more dots
> or fiddling with the handles to reshape the line, you are going to waste lots
> of time.
I must point out that for simply-shaped selections, adjusting the path
handles gives results of higher quality, faster.
>The trick with Freehand select is to do a rough selection and then do
the precise work using small closed strokes to add or subtract onto
the selection. The feedback is immediate, and is easy to draw a
precise line with the mouse as long as it is short.
Yes, agreed on all points here. Try using freehand select with a
tablet though (if anyone you know has one you can borrow) -- it
eliminates some of the complexities that you are describing, so that
you can select an entire complex shape accurately in one go.
> - The selection using a path matches the line in the path, which may be what
> we are referring to as precise, but this is actually due to the path
> selection being less flexible than a Freehand selection.
As i pointed out above, the freehand tool is also less flexible than
the path tool.
> In a Freehand select you can get a selection that matches the line you draw,
> just like the path select, if you turn of the "Feather edges", but instead of
> being precise, it is precisely what I don't want. When you select people in
> your image without feathering the edge and paste them back on a background
> that has been blurred, they have a cut out with an exacto knife look. There
> may be a way to get feathered edges and antialiasing with the path tool, but
> this is still a problem because you can't see what you actually selected.
> The actual selection depends on the amount of feather (radius), the radius of
> curvature (shape) of the selection line.
> - The feedback from a freehand select is immediate. You can see what your
> selection is as you make it without having to wait until after you have
> completed a long path description. If you are able to feather the edges on a
> path, most likely with a time consuming post processing, it won't be the
> precise edge you described with the path. It might not fully include
> everything you want.
EXACTLY! and no selection tool actually supports the function you are
describing -- the selection outline you see is unfeathered until the
selection is finished.
And the postprocessing is exactly as time-consuming as it is for the
freehand tool when you turn on the feathering option -- because it's
exactly the same postprocessing, the only difference is that it
happens automatically for freehand-select. (select->feather)
> - In my opinion the freehand tool is more intuitive for all but the very
> experienced user. Making the freehand select work the way most user's would
> expect would lower the learning curve for new users. The idea of adding to,
> subtracting from, or intersecting with the current selection is fairly
> straightforward and powerful.
Yes, and is already supported. You need a reliable way to reproduce
this; otherwise, people will just assume you are using an old version
of GIMP, and tell you to upgrade.
> - The fact that select by color supports the same booleans, feathering, and
> can be used in conjunction with freehand makes an even more powerful
> combination. Using select by color to differentiate large regions and then
> the freehand to cut or add to the selection is real efficient (it is also
> where the problem with the mode change when you are over a selected region is
> most evident).
Of course you can use select by color in conjunction with freehand;
the same is true for any other two selection tools, including path
(rightclick in paths dialog-> choose 'path to sel' /'add to sel'/
'subtract from sel'/'intersect with sel').
> >I'd reccomend you this workflow for background extraction:
> I like the methodology that I am using. The enhancement I am requesting
> would make my methodology much better, but even without the fix, I don't see
> myself reverting to path based selection for the majority of my work. It is
> not flexible enough and takes too much time.
Agreed (only!) on the fact that it takes too much time. It's more
flexible but harder to detail with (which is why using some
combination of tools is often wise -- if the shape you want to select
is simple, it can probably be done faster and better with path tool,
otherwise faster with freehand.)
It is true for the case of extracting a couple from the background,
that you could usually get *extremely* quick results using SIOX
foreground extraction, with only some minor feathering and touchup
with the freehand selection tool needed afterwards.
> >Your other proposal (selection stack) is very interesting, but sounds
> >quite difficult tu implement.
> >I'm not thinking exactly in your examples, but in the possibility to
> >transform different selection levels independently (i.e. you selected 3
> >circles and you want to adjust the size of the first one).
> If it is difficult then it is probably a non-starter. I viewed it as a "nice
> to have" feature and thought that it was worth brining up in case it wasn't
> too hard to implement. I thought that since the current behavior acts a
> little like a single layer stack, there was a chance that adding more depth
> might be doable without a huge effort.
I think that 'difficult' is the wrong phrase. 'already exists' is
probably the right one.
See the Channels dialog -- that's exactly what it is for.
Gimp-developer mailing list