W.I.P. Lime Rock

[Pixelchaser]

will be working on this alongside other projects. this one being the attempted use of publicly available .Las data.

and my far scale low poly outer scenery so far.

 

[LilSKi]

Very interested in this. It was a toss up if i was going to do Limerock or NJMP but I went with NJMP as I had actually driven it. Limerock deserves a solid AC rendition.

 

[Pixelchaser]

well hopefully I can do it justice. but that remains to be seen. for me this is literally a build to be done just like a kunos track. I think the outer scenery is kinda more apparent here than other tracks, but its just bloody trees everywhere .. its still a corridor style track build for the hi rez stuff.

really wanted to do barber though.but alas, limerock just seems perfect to be honest, nice and classic small solid track. bit heavy on the trees though, that's gona be tough.

whats out there for lime rock is quite rubbish looking. and even looking at the iracing version which I was fond of. geez its not that nice either generally. so looks wise ill make it better for sure. accuracy wise, how accurate do you reckon the iracing version was. was it even scanned ?

 

[luchian]

AFAIK, all iR tracks are scanned. And road surface rendition in iR is one of the best, if not THE best out there, imo. So I would already consider it a big achievement getting close to iR .
I do agree about scenery though, that is usually pretty weak in iR, so more can be done. Given the size of the track, maybe even some 3d trees .

Anyways, great project, I'm almost envious that I can't tackle it . I like the start, looking forward to see where it goes.

 

[Pixelchaser]

I built the basic layout in rtb last night and drove it . its my outer scenery anyway. but strangely enough after watching videos. I felt I already made a reasonable version to be honest bar plenty undulation issues. but that's where the laser data will rock, there will be no reasonable, just accuracy. and well tbh it must be really hard to go wrong with it pending the skills required to locate and model properly.

 

[Pixelchaser]

so Lilski. are you centre line splining ? or do you spline the edges then allocate the polys/verts ? the track widths on this track are all over the place. but ive not done that edge layouts before.

 

[LilSKi]

So far all my stuff has been center line spline. The only thing I did an edge spline with was the long 10* banked turn 9 for Riverside. But even the 6* banked final turn 9 at NJMP I did center line. It is easy when you have data to line up with. Riverside had practically no data so required a different approach.

 

[Pixelchaser]

thank you. ok I feel plenty hours excessive camber work can be negated by working with the edge you see. just not sure.

 

[LilSKi]

I think the layout for NJMP took me a total of 3 hours time. BUT I think blenders spline functionality is superior to 3DS max for track building.

In comparison Riverside was about 3 or 4 months to sort out the layout thanks to the lack of data.

 

[Portland12345]

Nice.

Good luck on project.


There is another track that seems to have LIDAR data

Summit point raceway.

 

[LilSKi]

Nice.

Good luck on project.


There is another track that seems to have LIDAR data

Summit point raceway.

There are many eastern US tracks that have lidar data. Just only so many people to make them. NJMP is just at the 4 month point since I started. These things take time if you want them to look better than something from 1999.

I have considered Summit Point with all of its tracks but I'm pretty well set on Watkins Glen after NJMP and an update to Bridgehampton so it will be a while.

 

[Pixelchaser]

I`ve decided to go down the nurbs modeling route. converting splines to nurbs this way I can track the exact width changes.

max has this utterly amazing function.the "pro Optimiser" its effectively a lod-ing tool. but its super powerfull. can keep all edges, borders, and even textures in tact. I will be able to use this las mesh directly it seems. to retain all the form and hand tune later on. I try demo it with a gif. stats on right hand side. 10% (down to 2% with no loss of "form".

another aspect is given Lime Rock is completely surrounded by trees. I may lay trees and tree lines and delete any hi rez terrain underneath in favour of a very simple form. this could leave even more of this las data form to be used close by with some smoothing where fit of course. I may even grab more las data for the surroundings given this functionality.

 

[Pixelchaser]

thinking about how to tackle the necessity of superb trees and im going to try somthing crazy.

track length, 2410 metres. lets say I had a 10 metre wide 3d tree every where around the track outer ring. that would be 241 trees per side to cover the side (not that I would) and the inner ring 482 trees. which we would not need. and considering much of the track is not surrounded by trees I think 300 sounds good. so 300 4000 poly trees. if I did an autumn version (90% leaves on the ground) I could pull this off I think.. id be prepared to create atleast 20 variation including the few evergreens. . from far side to far side most of the track obscures itself also and that would be extra good for some loding, crazy ?

 

[LilSKi]

IMO 3D trees are not the way to go. At this stage I feel a proper Y tree looks better than a 3D tree. And you can't really afford self shadows on the 3D trees so they just don't look right. The only 3D tree to ever look good IMO is the VIR tree. And it is WAY to complex for more than one tree. And it could murder performance when you get close to it.

 

[Pixelchaser]

I'm inclined to agree when sticking to the usual shader set but I'm sure it can be done quite well other ways too. we shall see.

great progress on my lidar meshes, here we have 41 million optimised down to 200K and 2 million down to 35K for the internal mesh, ohh yeah happy happy. and big thanks to Lilski for donating the links to the data. cheers fella !

 

[luchian]

Man, this thread starts feeling like being outside a candy store licking the window, while a happy fatty enjoys all that chocolate inside . Looks good Ali, keep it up !

 

[Pixelchaser]

tbh, im presuming theres a catch I haven't stumbled upon yet.

 

[Pixelchaser]

not to shabby for a first look in the editor with basic shader. fantastic resource to work from. have used cameras and correlated with views of the scenery from video and its definitely good enough.

 

[Mr Whippy]

I've found pro-optimiser ok for distant terrain optimisation.

My 2p, render out a normal map off the HQ mesh, projected off the optimised mesh) and that *should* add some nice quality back to the distance terrain when the sunlight gets low and the shadows start to show up on mountain sides etc.
If you have say 4 geometries (north, south, east, west) and 4 textures, you can then cut the batches and VRAM use by about half.
That lets you either use more geometry or pixel density.


There are many tricks you can do with this entire process.

There is a script out there which takes a camera and does a front/backface check, and selects front faces.

If you animate the camera along the course spline roughly, you can add all the back-faces to a bin-list and delete them.
Oooorr, you can select them and apply a much higher optimisation amount to those areas, since you'll almost never see them.


Also if you're really wanting the very best results, I think (I've not tested this one, but just thought of it now), you could select the border edges between front/back face selections in the above process, and then reselect faces from that edge strip.
That would define the horizon/silhoutte, which you could keep nice and crisp. You could then invert the face selection and optimise that down more.
You could then use normal maps to fill in the missing details from a more aggressive optimisation.


I've created tracks using this approach initially. Then I've created my 'tidy' mesh over the top for the track/tracksides/infield terrain, then I 'conform' the mesh to the HQ terrain.

That way you get nice flowing poly work that is easy to adjust, cut into, etc, rather than an arbitrary triangle mesh with random poly flow, thin triangles, etc.




On 3D trees, I don't see what is wrong with 3D trees in AC if they're done well.

The ones you see in the 3D trees/vertex normals/normal thief examples on polycount.net are nice for example... it just needs people to make them which I'll agree isn't trivial!

Just making a half decent scratch tree set (textures and geometry and normals) will be several hundred hours!

Dave

 

[Pixelchaser]

Oooorr, you can select them and apply a much higher optimisation amount to those areas, since you'll almost never see them.

already manually done.

Speed Tree speeds it up .

in a normal game engine yes that's all possible. and regards trees, the problem is inherent within the graphics engine. not everything is actually possible in this game engine contrary to the myth that all processes can work same here , its simply not true. there is some issue somewhere which leads to trees and grass shaders and all wide spread object having centre point issues and relating to the light.