Log in

View Full Version : Fusion - pyramidal image processing for video [experimental]


Pages : 1 [2]

ajp_anton
24th September 2012, 20:04
Thanks, now it works.
But if this shouldn't happen, simply clamping the output feels like an ugly workaround for a bug in the algorithm. Or maybe that's how it's supposed to work?

wonkey_monkey
24th September 2012, 20:21
But if this shouldn't happen

It's not really a bug, just an undesirable consequence of the maths. Now with an undesirable fix ;)

David

ajp_anton
5th March 2013, 06:15
I've used this a lot recently for photos since I just like its results better than anything else.
However, using it for (hi-res) panoramas is quite taxing on RAM. Could you make a 64-bit version of this?

wonkey_monkey
5th March 2013, 19:09
I've never compiled a 64 bit plugin before, but this seems to work:

http://horman.net/fusionx64.zip

Avisynth isn't really an ideal way to be fusing panoramas, though, and one day I intend to write a fuser to complement multiblend (http://horman.net/multiblend/).

David

ajp_anton
6th March 2013, 00:43
A replacement for enfuse used by Hugin would be great, as that's the program/GUI I use to stitch the panoramas together into two different exposures, as well as align single bracketed photos.
I don't like enfuse's results, so I always import the aligned photos/panos into Avisynth for this plugin.

The 64-bit version works, thanks. 64-bit Avisynth is a bit buggy though when I try to preview the fuse in lower res, but at full res it's fine.

ajp_anton
1st May 2013, 10:11
Another weird behavior that I'm wondering about...

Threshold set to
1.00 (http://ajpanton.se/1.00.png), 0.70 (http://ajpanton.se/0.70.png), 0.65 (http://ajpanton.se/0.65.png), 0.00 (http://ajpanton.se/0.00.png)
(The 0.00 and 1.00 are basically just the original images.)

You can see that when the sky starts to fill up, it does so in segments and leaves noticeable edges, and the fact that they are so grid-like with straight lines and straight edges makes them look very artificial. Can this be avoided?

wonkey_monkey
2nd May 2013, 12:31
Fusion cheats a little by using a very simple, very fast, but discontinuous function for downsizing/upsizing - I think what you're seeing is the result of that. Enfuse uses a slower but continuous function, so probably wouldn't exhibit this effect (or not to this degree).

Limiting the number of levels might help, but I don't seem to have included that parameter during the rewrite. You could try the old version (http://horman.net/fusion.zip) for now.

Reel.Deel
23rd November 2014, 22:30
@davidhorman

I'm just wondering what's the recommended (latest) version?

fusion.zip is dated 2012-09-24
fusion2.zip is dated 2011-11-31

The first post says that fusion2.zip is the latest version? :confused: Is that correct?

wonkey_monkey
24th November 2014, 19:13
It looks like I overwrote fusion.zip when I wrote this post (http://forum.doom9.org/showthread.php?p=1592727#post1592727), so that's the one to use. I've deleted fusion2.zip.

Reel.Deel
24th November 2014, 19:29
Ok, thanks. Out of curiosity what does the clamp really parameter do? I really can't see a difference between true or false.

wonkey_monkey
25th November 2014, 10:08
It keeps each pixel value within the range of the two pixel values from each clip. It won't do anything if the pixel value is already between the two, which it usually is.