Snow system

[FF.Fan]

Hi,

I'm trying to reproduce the effect of snow deformation on the ground as seen in this video at second 30" (snow battlefield in Virtua Fighter 4)

I managed to get what i want, I use the mesh api to access each vertex and calculate everything, but the problem is as soon as I increase the size of the terrain (bigger like more than 32*32), of course, everything get's laggy as I go through each pixel, every frame ! lol ...

Question is : have you got some idea or other method to improve the speed ?

I hope I will not have to stick with small area snow in a scene
Here is a screen of the snow implemented in the sample of shadow detection i provided earlier :

[TRANE23]

So your problem is the Vertex access on a surface with a lot of Poly. Try using a chunks system(with size < 32*32) by partitioning your space, detect the current chunk(with sensor or collider) and do the job with mesh api only on this one.

[FF.Fan]

So your problem is the Vertex access on a surface with a lot of Poly. Try using a chunks system(with size < 32*32) by partitioning your space, detect the current chunk(with sensor or collider) and do the job with mesh api only on this one.

thanks, partitionning in chunk is indeed a good idea ! i'll try it.

When i woke up this morning I got an idea, as my problem is to know which vertex is concerned without going through all vertices of the poly, then, as the snow area doesnt move too, what about precompute in a big hastable all the vertices position and then i just need to pick the one closest on the hashtable and do my thing.

I'll try both method and see what's working.

Thanks again.

[TRANE23]

I think it will be slower, you'll have to test all vertices in the hashtable and I'm pretty sure the access via mesh API will be faster than via hashtable(if it's a big one).
A more complex way would be using a space data structure to store the vertex like a BSP or an R-Tree which will allow you to discard vertices for the "closest test" but imo the chunks and the sensor will give you approximately the same FPS result and is much easier to code.
You could also mixing the two methods : bigger chunks with Partition Space Data Structure to discard vertices during the test .

[FF.Fan]

I think it will be slower, you'll have to test all vertices in the hashtable and I'm pretty sure the access via mesh API will be faster than via hashtable(if it's a big one).
A more complex way would be using a space data structure to store the vertex like a BSP or an R-Tree which will allow you to discard vertices for the "closest test" but imo the chunks and the sensor will give you approximately the same FPS result and is much easier to code.
You could also mixing the two methods : bigger chunks with Partition Space Data Structure to discard vertices during the test .

I think you are totally right about that, its indeed the same if I use a big hashtable.
I try the chunks & sensor approach.

I really don't know anything about R-Tree, BSP and other. I looked on Wikipedia but its look kind of tedious to implement ^^

Thanks a lot anyway !

[FF.Fan]

Yay ! I finally got the chunked system working ! its nice an smooth with big area now
Thanks for the advice.

the only problem with that technique is, as seen in this screenshot, the material.


Indead, as its a separated mesh, it doesnt shade the same. Is there another trick ?

[TRANE23]

You have to set the same normal for the common vertices

[FF.Fan]

i'll try, thank you again !

[FF.Fan]

Well as each chunk is a separated mesh, i can't really manage to get the overlapped vertices and set them to the same normal
any hints ?
thanks

[NiCoX]

Hi,

There are many solutions, but perhaps the easiest at this point of the development (to not rewrite all) would be a "polish" post pass: for a given "frame", once you have modified the affected chunks using your existing algorithm, you just have to run an average filter on the (modified) borders normals, taking into account, obviously, the adjacent chunks. And the same (and in the same time/pass) for positions, welding vertices to avoid holes. This can be pretty fast as you'll only work on a limited set of points.

[Deivid L.]

wow! thanks brother
This is very useful.

[FF.Fan]

Thanks for the tips, unfortunately i can't test right now, maybe tonight, if I manage to do what i want, I'll post the code sample for everyone

[FF.Fan]

I really can't make it, if somebody is kind enough to explain me :
here is the source : https://docs.google.com/open?id=0BzzqHewTzXc-X1I0bnNmdVY2d1E
thanks

[TRANE23]

Check your PM

[juaxix]

I have tested the code, it is a nice idea, did you improve this?