Several questions, but basically, is X a problem?

Summary:

A reply to the question of this problem with other implementations would give a lot of insight as to this one. The problem: not widely distributed energylines (I think)

Several questions, but basically, is X a problem?

Anonymous (70.91.214.x) 23 Oct 2007 19:31

Okay, apologies in advance for the high density of unknown information.

I read of a problem that some implementations of this technique have (YouTube video id b39R8eHyvmw), and I'm wondering if Liquid Rescale has sidestepped this problem, or if it should become a feature request.

Here's the text:
(the video's program appears to not…)
"get hung up in any spots, and continuously remove from the same seams.
I've seen a few people try to write their own image carving before… and they usually make the mistake of 'real time' carving, which generally gets hung up on the same seams over and over.
Do you use the seam ordering method described in the original idea? (Caching seam pixels in order of removal, before removing any seams.)"

My eyeball of one video that may not be of the same program, (I'm not sure, check yt id TDZc3HnKp8o says "yes, this is a problem" but I'm a newb to the whole subject.

It may also only appear to be a problem in the video due to image features (painted line) that make it easy to detect distortions, and the use of the "delete" option may bias/require the lines removed to be near that object for awhile.

I only found a few videos of the Gimp plugin, and none of them in real time like this vid of the original method: vIFCV2spKtg.

And lastly, video questions. It doesn't seem possible yet to use a GIMP plugin with VirtualDub, and it sounds so hard I won't ask. Can this plugin work in batch mode? And, most-highly-speculative-question is, is this technique (find lines of least energy) feasible in reducing video filesize?

Re: Several questions, but basically, is X a problem?

CarloB 23 Nov 2007 16:42

First, sorry for the delay…

About real-time scaling and such: the Liquid Rescale implementation of the algorithm does support it (in theory): it builds a visibility map and uses it to decide which pixels to show, and of course also to decide where to insert seams for enlarging.
So the anser is: yes, it avoids enlarging every time in the same place if possible.

The video TDZc3HnKp8o is about another program I don't know anything about, but the explanations you proposed seam quite reasonable.

About working in batch mode : it can, from inside the GIMP. I wish to split the code at some point and make an independent library, but I don't know when.

Video filesize: I think there are far better compression methods, which avoid distorting the image! If you were wandering about aspect ratio instead, it is possible in principle but I wouldn't count much on the quality of the results … unless you want to create some kind of liquid effect in a video (that would be interesting).