Oh, like Speedtree Live.[1] Speedtree has been around since 2000, and it's pretty good. I used to keep the Speedtree Million Tree Forest demo executable around to run for relaxation.
Epic has been developing in-house replacements for middleware used with Unreal Engine. Metahuman replaces other character builders. UE has replaced Simplygon with a new level of detail generator. The PhysX physics engine was replaced with UE's own Chaos Physics.
Now it looks like Speedtree is getting replaced.
Probably because the IDV, the company behind Speedtree, was acquired by Unity in 2021. Simplygon was acquired by Microsoft. PhysX was acquired by NVidia. Epic supports UE5 on a wide array of competing platforms, and they don't want to be dependent on middleware that's owned by a platform or a competitor. It's bad for their business model.
Unreal Engine has an unusual business model. It's free to use until you get a million dollars in revenue. Then they charge a percentage of revenue. This reflects the economics of the game industry. Most of the profits come from the top 10 or 20 games. So the number of customers they actually have to bill is quite small.
For anyone interested to look more in-depth on the subject, there's an entire book dedicated to it called Algorithmic Beauty of Plants. The math is really cool!
ML systems are more difficult to tweak. E.g. if you want to adjust parameters such as growth speed, branching factor, sunlight attraction, etc., and see how the resulting plant changes.
Sure some NN tool can ape anything however that approach sheds zero intuition on understanding the hidden structure of what it's mimicking... current wave of NN masks asking the important questions like how does a cell grow ... I fear NN mania kicks the can down the road without revealing anything fundamental to understanding
In general we know how cells grow, and even plants. Actual plants are controlled by variables that vary species to species, can't be easily measured, and are too numerous to count.
There are a lot of unanswered questions related to how plants grow. If you look through the recent papers at Algorithmic Botany (linked at the start of the thread), you'll notice that they're partnering with other labs to do fundamental biological research (specifically, on the growth of veins in flowers and leaves). That's because they reached the limit of existing knowledge and needed to learn more to improve the computer models.
That's where I did my master's, though I didn't meet Holly and Steve. I'm told they left for Weta shortly before I joined.
Google published some work on reaction-diffusion systems implemented as a ML system. If you would put that on steroids, you could learn and simulate the growth of complex organisms I think.
This looks awesome, could be super helpful for synthetic data generation, for a lot of industries including mine (drones). The UE compatibility is a very good selling point, and the procedural nature is also a good selling point (less overfitting of neural nets trained on a limited set of tree models)
Thanks for this! For the last couple years I've occasionally wondered "What was that tree rendering startup in YC a few years back?" and could never find it. Was starting to think I'd imagined it.
Hi! Nice to see you on here! Yes that is an interesting application for synthetic data generation that we plan to look into - I will contact you when we head down that path!
FWIW, Many people use UE for synthetic data generation, so a plugin that generates trees to add in UE scenes is already super useful for that use case.
I made something similar for the weird Lovecraftian text editor I'm making. My goal's to eventually let people fill a wasteland with flora by doing creative writing around it. Mine's more stylized, simpler, and Unity. :p
I'd love to see this as part of a permaculture planning tool. Though it would probably require a lot more different kinds of growth parameters.
Figs and bayans, for example, won't grow like the typical hardwood deciduous tree ... while most trees grows with some kind of symmetry between upper limbs and the root system, fig trees don't follow that.
The potential applications to permaculture planning are quite exciting. We can leverage the basic aspects of our current plant models and create new models for different species that capture not only their appearance but their behavior to different environmental factors.
The upper limbs of a fig tree seem to grow whichever way to reach the light, and the root system will also grow like that. A fig tree might be trimmed and still have an extensive root system.
Hardwoods will try to maintain some kind of symmetry. One root that branches supports a branch above ground. If that root is cut, the above-soil branch it supports dies out. Likewise, cut the branch, and if it does not grow back, the root will die back.
This is why for hardwoods and most trees, when you transplant them bareroot, it’s best to spread the roots out.
> One root that branches supports a branch above ground. If that root is cut, the above-soil branch it supports dies out.
Do I get this right? You're implying that specific roots are somehow more or less associated with specific branches in most hardwood species but this association is less strong in fig trees?
And the association comes from capillaries running all the way from the roots to the twigs with basically no interconnections in between (that'd be in the stem)?
This is really cool. I would love a VR meditation-style app where you just planted trees and watched them grow. Perhaps set at the foot of Mount Fuji with some daily haikus or something.
We do simulate branch shedding for branches that are not doing too well. Branches can also be human pruned. Branches breaking due to weather or strain over time is an interesting addition to look into!
Epic has been developing in-house replacements for middleware used with Unreal Engine. Metahuman replaces other character builders. UE has replaced Simplygon with a new level of detail generator. The PhysX physics engine was replaced with UE's own Chaos Physics. Now it looks like Speedtree is getting replaced.
Probably because the IDV, the company behind Speedtree, was acquired by Unity in 2021. Simplygon was acquired by Microsoft. PhysX was acquired by NVidia. Epic supports UE5 on a wide array of competing platforms, and they don't want to be dependent on middleware that's owned by a platform or a competitor. It's bad for their business model.
Unreal Engine has an unusual business model. It's free to use until you get a million dollars in revenue. Then they charge a percentage of revenue. This reflects the economics of the game industry. Most of the profits come from the top 10 or 20 games. So the number of customers they actually have to bill is quite small.
[1] https://www.youtube.com/watch?v=p0DwVQA-4Ns