This is a couple of basic scenes that I have produced using my stitching algorithm. These images are based on 16 tiles in a 4x4 arrangement. I'd like to know what you think of the quality of the stitching.
[attachthumb=1]
This is a stitched image
[attachthumb=2]
This is the original scene
[attachthumb=3]
Another stitched imaged - there are a couple of stitching artefacts visible if you look closely
[attachthumb=4]
The original scene
Let me know what you think...
Nope, cant tell the difference, every time I think I've spotted an artifact I check and nope its in the render lol. I'd like to see some supper size images sometime :)
richard
ps I could send you one of my planets, loads of detail in there so lots of opportunity for artifacts to creep in, let me know but keep up with this, its going to VERY useful :)
The first image is nicely stitched, can't find any artefacts that quickly.
The second one has an obvious large seam through the water.
The atmosphere is a bit more hazy in that one and that's probably the reason why the tiles differ more.
Martin
The shadows are darker in the stitched on for some reason.
Openning these up in seperate tabs and flicking back and fourth between them, there's some pretty horrible differences. Which would concern me if this was used in an animation. I'm sure that's not the stitching purpose. Just pointing it out.
Ah yes with the tab trick I can see what you mean. I don't think this is due to the stitching algorithm or process but rather Terragens clipping and GI issues. Maybe the shadows could be sorted out by upping the "ray detail padding"and the trees by upping their render quality but both these will add considerably to render time.
Richard
You also might want to have a few clouds, as artifacts can come in theres stitching wise too.
Thanks guys for all the feedback - this is the sort of information I need, particularly if the render parameters are going to be have to changed so much that splitting the image up to render is not much quicker than rendering it as a single picture.
Quote from: Tangled-Universe on August 10, 2009, 03:20:02 PM
The first image is nicely stitched, can't find any artefacts that quickly.
The second one has an obvious large seam through the water.
The atmosphere is a bit more hazy in that one and that's probably the reason why the tiles differ more.
Martin
This is the worst stitch I can currently come up with - it's the image I've been tweaking the code with as I know it looks pretty bad in the lake area. It appears the more "busy" the scene in the area of the stitching, the less obvious the stitch is.
Quote from: Hetzen on August 10, 2009, 08:37:12 PM
Openning these up in seperate tabs and flicking back and fourth between them, there's some pretty horrible differences. Which would concern me if this was used in an animation. I'm sure that's not the stitching purpose. Just pointing it out.
I've been using the excellent Beyond Compare utility to compare my image outputs, and in the tree scene it appears the trees render a bit lighter in the stitched scene, but it is mostly a consistent change across the scene. If you were using the software for broadcast quality animation, I would suggest rendering each frame as a whole rather than tiling it (my software will support this eventually).
Quote from: neuspadrin on August 10, 2009, 08:59:25 PM
You also might want to have a few clouds, as artifacts can come in theres stitching wise too.
I have tried a couple of scenes with clouds and so far there are no problems.
I'm going to tweak my code a bit futher and I hope to produce one or two larger, higher quality renders over the next week or two.
Can you make it so that it automatically sets the ray_detail_region to "Detail in camera" (value of 2 in the XML file). That should eliminate differences in shadows cast by terrain. For GI differences (usually noticeable in atmosphere) you will need a larger overlap between tiles, but higher values for GI Sample Quality will make this less necessary.
Matt
Quote from: Matt on August 11, 2009, 12:36:16 AM
Can you make it so that it automatically sets the ray_detail_region to "Detail in camera" (value of 2 in the XML file). That should eliminate differences in shadows cast by terrain. For GI differences (usually noticeable in atmosphere) you will need a larger overlap between tiles, but higher values for GI Sample Quality will make this less necessary.
Matt
Matt, do you think it is feasible to completely solve this problem? Using fill lights and no GI just isn't an option for say cubemaps with clouds and the like.
Cheers,
Martin
Higher values for GI sample quality will give better results. I will work on other ideas to make things match even with low GI quality in future.
Matt
good to hear that !
it will be very cool to be able to stitch several crop renders without lighting problems ;)
and without the time killing GI :p
Quote from: Seth on August 11, 2009, 06:21:03 PM
good to hear that !
it will be very cool to be able to stitch several crop renders without lighting problems ;)
and without the time killing GI :p
The "fun" thing is that you can easily stitch renders together which do NOT have GI but use fill lights.
When you're using GI and have quite some clouds and haze for example you virtually can't stitch them at the moment.
Increasing the GI sampling helps, like Matt says, though not optimal.