Polygons, voxels, SDFs... what will our geometry be made of in the future?
- Thread starter eloic
- Start date
Ok, not the most appealing environment and it's quite monotonous, but holy crap! It's amazing how you can just copy/paste the same scultpure ad infinitum with no performance penalties. With a more refined artistic vision, this could be used to achieve even more impressive results, including microdetail.
Lin has finally released clips and information about his voxel water technology.
EDIT #1: Previous video and information:
EDIT #2:
EDIT #3:
John Lin said:It took a bit longer than expected, but the fluid simulation design from a few months ago has been implemented in our voxel project. It uses the same cell/particle hybrid method and is therefore fully volumetric with no height range limit beyond the 0-4095 y world boundary! It will flow through buildings, fall down into caves, flow over and under cliffs, fill up containers and lakebeds, and navigate through tunnels. The design has been adapted to run sparsely and inline with the world generation. The entire pipeline has also had its design improved to be 99% multithreaded, with a 1% critical section that has the smallest execution time. In this video, 4 CPU threads were used and the simulation time never exceeded 8ms.
Saving to/loading from disk, generating it naturally, spawning and controlling it, and not pouring into unloaded areas are implemented. Currently it only interacts with terrain and buildings as seen in the video, but player and object interactions will come soon. Some improvements to be made include the sound calculations, rendering and supplemental particles.
The video used faucets for the waterfalls that will run indefinitely (and eventually fill up the whole world), because there's no evaporative water cycle. With water being finite, waterfalls spawning from a body of water will eventually run out. I still don't have a good idea to solve this, but I'm definitely open to any ideas.
As always, thanks for all your support and patience!
EDIT #1: Previous video and information:
John Lin said:This was a research project that aimed to find a real-time realistic fluid simulation that was able to handle moderately complex scenes without taxing the GPU horribly or wiping out the entire CPU. It's a hybrid lagrangian-eulerian method based on the work by "Hu, Yuanming and Fang, Yu and Ge, Ziheng and Qu, Ziyin and Zhu, Yixin and Pradhana, Andre and Jiang, Chenfanfu" (2018). Special thanks to Grant Kot for his epic fluid simulations, for without him I never would've believed CPU fluid simulation was possible and tore my hair out trying to figure out how he made his stuff look so awesome. Give him a follow for even cooler water stuff: twitter.com/kotsoft
The entire core of the simulation, as well as the CPU-GPU transfers (which, sadly, actually have a big impact on frame times) is executed on the same main thread and is mostly accelerated using AVX2. The GPU is of course used to do all of the particle & scene rendering, and if it used multithreading the rendering could definitely appear smoother. It uses C++ and Vulkan, and was ran on a base i7-8700K and 2080 Ti.
Spoilers: This is going in the game I've been showcasing, hence not wanting to bottleneck the entire system. Expect to see this in a prettier environment soon!
EDIT #2:
John Lin said:
EDIT #3:
John Lin said:
Last edited:
Silly video, but look at this:
The ground is covered in "true" leaves, no flat texture. On a PS4.
The ground is covered in "true" leaves, no flat texture. On a PS4.
Same problem as I see in most of these vidoes...the leaves are "dead" (static)...which kills the suspension of disbelief in my brain.
Sadly yes...TW3 has made it very hard for me to play games with dead foilage...it is like AA or Raytracing...once you see it done right/better...it is impossible to go back.
Yes, it's a pity that we're getting better at so many things, but some other things seem to just stagnate, like fully interactable environments.
In this case, I guess a solution could be add just a few movable leaves, so that an effect is achieved.
At any rate, I like that the ground doesn't look flat and that you can appreciate true geometric detail there. Also, the sense of volume and depth in the bushes is great, they're not just a bunch of textured cards.
I get perhaps more "punished" by such things as I am a veteran and foliage is something I spent years observing...so dead static foliage really annoy my grey ones.
Great ligthing?
I think the lighting looks very "bloomed" and not very realistic at all.
I still don't find it "great"...but then again raytracing has spoiled me /shrugs
Wow, how lovely.
I still think it's that 'puppets on ropes' physics which limits Dreams the most. You can make nice kids / casual stuff, but no serious action game, sadly.
Also in this video the TAA lag is very noticeable, which was mentioned as an issue in the paper.
Very impressive otherwise.
um... can somebody explain me what's the motivation / application here? I really don't get it yet. Compression? Detail? Rendering?
Compression and better rendering performance are at least possible. Produced output is differentiable and also good for training other neural nets. If you go through twitter hooplaa related to this paper that would give good idea of one neural representation approach.
There is ton of twitter messages from casualeffects(click the the link). Also the authors of the paper have been very active on twitter.
you're gonna wish you had tensor cores for this.
It's just amazing how far rendering has come during my lifetime. I remember learning in early 90's to draw simple geometry, then flat shading, camera movement, gouraud, texture mapping, bump mapping. Implementing it all in purely motorola 680x0 assembler. And now we have freaking neural nets and what have you. Time has truly passed what I know about 3d graphics.