Thanks David, thanks Andy!

I tried again (after updating) on the same image (as -I suppose- you too did).
Output is certainly a lot better, still not as expected, as you will have noticed.
Some smooth transition between modified areas is probably required, as well as the removal of the shadows of the two animals, as requested by the red mask.

Anyhow, thanks for the quick improvement.

This is all I get:



No idea why

Ah yes pirate541, that's almost certainly a flaw in the filter (so not your PC don't worry!). As I say we're trying to figure out alternate methods for this, I'm already thinking it through

Thank You garagecoder

Pirate541
That was what I got.
I thought it was due to my old version.

Updated and the preview showed the changes and it worked well.

But I try the new option that adjusts the size of the preview and now I can not use more G'MIC.

Updated the G'MIC, but the screen is too big, and not re-scales to get smaller.

Can apply no effect, the save button is not visible.
Anyone have any tips how to solve this?

Thx

Hah oh dear Akros everything goes wrong today
I did read somebody else with the preview problem fixed it by changing preview to a small size, then minimized gimp. I hope this works for you!

Edit: you might also need to first open an image which is wider than tall.

Figured out how to re-use the G'MIC.
I opened another very small image, and therefore the preview was also small.
I was able so reverse the large size of the preview to normal.
And everything is back to normal.

Unfortunately, the Large option in the preview is impossible for me. But as I had never used this option before, this restriction will not feel much.

Now I can continue testing what's new.

Thx garagecoder!

Akros
The image I posted was after I updated Gmic

Hummm....
Before upgrading, I had the same kind of images with this filter.
With this version 1.5.9.1, the filter works without problems.

P.S:
Pirate541,
But your version is 1.5.8.3 , not much different from my previous version.
Does anyone with older versions can use this filter?

Hi David, hi Andy.
I post here the results of some other experiments made by me.
As you know, I consider this filter very useful, and I use in my image photo manipulations.
Because I noticed some "strange" behaviour when enlarging the photo by preserving primary subject, I decided to perform that function using a particular input, able to magnify the effects: a checkerboard.
For all my experiments I used an enlargement factor of 150% (width).
I selected a small area which I wanted to "preserve" (using green mask).
To distinguish clearly the area to be preserved I filled it witha different pattern.
As you can see from the attachments, the area is always well preserved, but the enlargement of the squares all around occurs with a really strange behaviour.
Particularly on the left side, and this irrespective from the fact that the area to be preserved be more centre-towards-right or centre-towards-left.
I attach here:
- input and output with preserved area towards right
- input and output with preserved area towards left (name have suffix L)
Everything in one folder, zipped.
Sorry to give you additional efforts, but I guess an amendment should be done on the implementation.
Thanks



PS: it seems the problems lies on the enlargement technique, even w/o areas to preserve

Actually with that type of image this is expected, because the gradient is almost constant the seams will simply be vertical over nearly the whole area (so there is nothing to choose one seam over another). I have thought about alternating seams which would at least produce symmetrical stretching, depending on how tricky it becomes I might be able to add that too. Thanks again for trying this out

OK, I understand, but I put that image to magnify the effect; something very similar occurs also with any of my "normal" images; the left side is "abnormally" enlarged, compared with the rest of the image

please, have a look at this new example (normal picture)





still the left side looks wrong

Hmm it's quite possible with some pictures this will always happen, due to an area of much lower gradient.

Anyway, I've made a small adjustment so that removal alternates where gradient is constant... this should be available to try under "testing->garagecoder->seam carve [fast]" but be aware this is a stripped down version with no support for priority mask or anti-aliasing - it's only there to test the checkerboard problem.

Hi Andy.
I tried you updated version (under test) for the enlarging mechanism.
It looks better, but there are still (serious?) problems: look at the church-steeple.
I hope you and David will sooner or later provide a version usable for actual photo manipulation.
Anyhow, thanks a lot for your support.

David, Andy,

is any activity going on on this filter or is it still in the same situation as a week ago?
More generally:
- do you plan to improve it to avoid/limit the unpleasant behavior in some cases?
-or
- will you leave as it is for a (maybe long) while because busy with other activities/duties?

(in the last case, imo it's better to move it from the list of "regular" filters -"Deformations"- and put it under "Testing", what do you think?)

I've been trying things out but really not satisfied with the output... using a more exact marking of seams, the removed/retained areas are precisely those selected (which is closer to what you might expect) but because high gradient areas tend to have vertical seams there are noticeable edges left over. It might require some inventive blending.

The alternating seam removal could be merged with the main filter at least, I see no reason for making it a test only filter as it still serves the original purpose of removing low gradient areas in a particular type of image.

But yes, you will have to be patient because I don't think it's a quick fix

...I'm patient
Thanks