Here is a short walk over the same area:
There are a few issues in this video. Some of the LOD switches are too visible. There are some noticeable discontinuities in the lighting that show in some distant cells. I'm working on fixing these issues, hopefully you can see past them.
Awesome stuff as usual.
ReplyDeleteDo you think the ground is a little too noisy? It seems like what appears to be sand should be a fair bit smoother.
Everything is a little too noisy. Remember this taken from a prototype that is mainly about the streaming. I'm taking care of the texturing in a different application. I have not posted anything about it yet.
Deletewhat is the data structure for storing the vox data(ps i am not a programmer so if you reply be simple) is it in an octree? is this vid also steamed over http?
ReplyDeleteI use an octree to store sections of terrain. These sections are conventional polygonal models. I download these over HTTP as the camera moves.
DeleteSo the paper you link to for your new example-based technique is one I've read before, and it's fascinating. The results they achieve are absolutely stunning, and it looks like you're getting comparable results.
ReplyDeleteI'm a bit confused, though, because they're simply using heightmaps, and you're using density map isosurfaces. It's easy to imagine how the same basic technique could be applied to 3D data, except that the sample data to draw from would be hard to come by. You could just convert the heightmaps into density maps, but the output you'd get from that would be roughly equivalent to the original heightmaps, without any overhangs or recesses, or the rich texturing of the near-vertical surfaces.
So my question is, what did you use for sample data?
I am not using the example base technique to generate large terrain features. They come from a initial noise and some filters with some global processing phases added later, like fluvial erosion. I am using samples to refine the cliff sides, rocks and pretty much every surface you see since the initial phase produces no detail at all.
DeleteAh. Which explains why the videos and scrollshots are focused on close shots of the rock and sand formations, not far-out shots of the macro features. I am VERY impressed with the way you've got the small peaks of sand with bits of rock protruding in the valleys between them. I just got back from a vacation in Palm Springs, and that part, at least, looks exactly like much of that desert.
DeleteThough I'm still curious where the samples come from. Did you hand make them yourself? Extrapolate them somehow from photos? Work from polygonal models from skilled artists?
Most of the samples are 2D textures I found on the web.
DeleteThat's got a very nice sense of space...it just feels _big_, in a way you don't often get in this sort of thing. All the more impressive since it is all explorable and not a skybox.
ReplyDeleteI wonder what it looks like forested.
You wouldnt be able to see far probably, but standing on one of those cliffs, or peeking through the foliage would be impressive I bet =3.
DeleteSo, this is lit with just the global light and your radiosity solution? Like the forest vids? The shading within the crags (and even small overhangs) of the cliff sides is spot on, in my opinion, so good show!
ReplyDeleteThat broad cliff side is just aching to have cliff dwellings in them, ala (http://photos.igougo.com/pictures-photos-p360727-Cliff_Dwellings_in_Petra.html).
Of course, that's a pretty specific type of "city", and might just be too far afield for procedural rules, this time around :)
The work is coming along great, BTW!
Thanks.
DeleteYes it is the same radiosity solution. There is no other source of illumination in these screenshots and videos. This makes even the direct sun shadows very soft. This is OK for clouded skies, for a desert scene not so much.