That's Rainier there on the horizon just to the right of Mt Hood ;)
Two terrains, 10m res centred on Mt Hood and 100m res centred between Rainier, Adams and St Helens. Single tree population, 2.6 million trees (within the frame), distribution controlled by a USGS canopy image map and restricted to 35km from the camera... dark green shader masked with the same image map to cover the terrain beyond 35km.
Slowly upping the detail in the shaders, although the main one was green grass which you can't actualy see in this render ;)
Also attached a screengrab of the node network which shows how my masking approach uses several components which are recombined to form multiple masks for different shaders.
Detail 1
AA 4
That's some pretty intense masking ;). The scene is looking good.
Wow, BigBen. Excellent work. Now I know why I have trouble following your explanations. Amazing work.
beautiful work here man!....
Quote from: calico on September 09, 2007, 10:49:58 AM
Wow, BigBen. Excellent work. Now I know why I have trouble following your explanations. Amazing work.
It's not really all that complicated (in TG). Like most surfacing/functions the key is first figuring out what you are trying to achieve. In this case the problem is populating trees on a terrain.
- How can I place them realistically - Image map of real data
- Why render trees at great distances when they can't be resolved as discrete objects - Distance shader based on render camera, coloured shader for terrain
- I want to distribute different age groups more realistically - Density of image map : Age of trees (more young trees at low density parts of image)
- I want to compensate for altitude changes affecting tree height - More younger trees at higher altitudes (not entirely correct, but younger = smaller)
- Slope orientation also plays a part in tree density (amount of sunlight/ evaporation, shaded side usually = higher density, lower altitude limit) - Slope face masks
- Some species are found more towards the bottom of gullies, or up higher along ridges - Image map of WM flow map
- Only populating objects that occur within the field of view saves a lot of RAM - FOV mask
- Trees don't grow on lakes/rivers (or only in shallow water) - Lake/river mask
So you have 1 shader to colour distance terrains and 1 or 2 tree species with 3 different age groups (young, middle, adult) and an image map that forms the primary distribution. You could create a mask for each object population, but it's more efficient to identify relationships between the distributions and then build components that can be shared between them. Once you have defined the relationships the rest is like putting lego blocks together. This is definitely one of the strengths of TG2.
Cool. Crazy large scales going on here. This is an example of what you can't do with procedurals, at least not yet.
I only worked with trees once but how do we get rid of the grainy look with vegetation that you always see? Does it just require very high render quality settings?
Quote from: efflux on September 09, 2007, 07:26:02 PM
Cool. Crazy large scales going on here. This is an example of what you can't do with procedurals, at least not yet.
I only worked with trees once but how do we get rid of the grainy look with vegetation that you always see? Does it just require very high render quality settings?
It depends on how far away the trees are from the camera (I'm assuming you're not referring to the "noise" in this image which is created by shadows of the objects which is the type of noise you want)... There is a render quality setting in the Population node. Increasing the quality for close up vegetation will improve the render at the cost of increased render times.
I see graininess whenever people use vegetation which obviously has a lot of rough detail. It's an issue with how TG2 handles things though. I haven't experimented so I don't know how to deal with it. I did one render with only a few trees and bumped the quality up really high. It didn't look too bad but I dread to think how long it would have taken to render if I'd had forests. Your image here is not that bad though and you're not at a stage where you'd want to bump up render settings anyway since it's a work in progress.
I have a few tests left to run on populations, but from some previous renders I've found setting the population render quality to the highest setting gives me results that I like. I agree that noise in tree populations can sometimes detract from a render, especially when the other surfacing looks stunning.
I'm currently running a test animation to figure out the soft shadow settings for my fill light setup (placing an "ambient" shadow under trees), and after that I'll be testing the render quality settings in populations to see how each affects the appearance of objects over different distances (and also memory usage). I used to set up duplicate populations with different render quality settings for different distances and I just want to see whether that approach was really necessary before I set up the rest of my trees.
[edit] - the other thing that helps a lot is getting some better lighting effects on leaves using translucency and reflectivity. This gives more realistic variations of leaf colours based on their position within the canopy.
This looks great Ben, especially the amount of trees and scale. I am *still* struggling with appropriate scale in my own St. Helens scene. Very frustrating.
Did you use billboards for the trees? If not this is a tremendous amount of geometry both before and after culling. Maybe worth reporting. It will show if you use the CLI version. I'd be interested in the number. ;D
- Oshyan
They're all billboards. I forgot to keep the log file (I only have a (small) collection of crashed logs). My next test will be from ground level so I'll see how far I can push the populator. I think I've pushed full objects to 20km with masking (1.5 million trees) previously but that will have to wait and see. Having watched the CLI for my high res panorama test, there are certainly some serious numbers involved and it will be interesting to see how the update goes.
As for scale, I'm using cbalaskas' billboard tgo http://forums.planetside.co.uk/index.php?topic=1477.msg14731#msg14731 (http://forums.planetside.co.uk/index.php?topic=1477.msg14731#msg14731) which is 10m high. The images are from XFrog (I bought myself some trees (NW USA) for my birthday :)), so I just use the dimensions mentioned in the accompanying PDF to scale the TGO. This should get a bit better with some additional masking planned to increase the porportion of younger trees at the edges of the distribution. Too many lone adult trees out in the open seems to screw up the scaling (especially near the snow line).
Quote from: Oshyan on September 10, 2007, 12:58:05 AM
This looks great Ben, especially the amount of trees and scale. I am *still* struggling with appropriate scale in my own St. Helens scene. Very frustrating.
Just cleaning up my data files and I noticed that your 3m St Helens terrain has UTM in the file name :) If you have a record of the coordinates of it it would be great, but I'll try and manually calculate the offset within my terrain set anyway and see how well it fits in. Should be interesting.
Yeah, not sure if the descriptions made it over to the new Ashundar site yet (soon at least), but I detailed a decent amount of info, including that I had converted to UTM. If I can dig up the original files (ummm...) I'll try to let you know coordinates.
- Oshyan
Don't worry too much about it. I should be able to get close enough with coordinates from the preview window and then manual tweaking of placement. Another good thing about being able to mask terrains is that I can crop out the flat bit of terrain on one edge with the missing data :)