Could AI be used to "upscale" BOTF's graphics?

[Deimos]

I notice a few old games from the early 2000's are being upscaled to "HD" with better looking graphics due to advances in AI software.

One of my old favourite games Settlers IV has had this done via an unofficial HD patch and it looks so much nicer.

E.G - https://settlers4-hd.com/

Could something like this be applied to BOTF in some way? at the very least, the 2D UI elements to make it look better and easier to read on larger screens.

[Tethys]

AI can actually do a lot of neat things. We used it over the last year to lightly build some supporting code for some Bridge Commander mods. It can also disassemble, however, with a grain of salt. Sometimes more often than not, things are amiss and information is erroneous. But the general information is there and can be extrapolated with a little less work than doing it all from scratch.

In short: YES you can probably have AI upsample the basic graphics of BOTF into any resolution you'd like, but then you still have to do the work of editing all the wdf files. That is something an AI would have to learn and get good at (meticulously changing all variable to match new graphics files)

Edit: I had a look at your link and that is such a MASSIVE graphical improvement!

[LadyGray]

I was about to make a post about this.


NVIDIA have things called DLSS look at how they improved Minecraft https://www.youtube.com/watch?v=Vgu6pLYNrPE


Yes I hope this becomes a priority so we can have good detailed high resolution combat scences soon.

[Tethys]

We used AI to help disassemble an exe. For the most part, it couldnt decide what one function could do, unless you tell it to analyze it again and add in other code. At a certain point, the AI fails to keep up and data is lost. The AI just isnt there yet in terms of functionality. I would much rather have an AI that could give us a working and detailed disassembly from which we can extrapolate a source..

[Flocke]

NVIDIA have things called DLSS look at how they improved Minecraft https://www.youtube.com/watch?v=Vgu6pLYNrPE

Sounds like magic, but I had a quick look what it does and found it is used to improve raytracing render speed.

When you know how computive raytracing is, it makes absolutely sense to use such an upsampling method to speed the rendering. That it uses AI technology to visually improve the upsampling is a logic choice, but it doesn't make it faster than traditional rasterized rendering. And it doesn't add generated objects for missing detail, it just helps to determine the proper color of in-between pixels, evaluating changes over time and possibly taking movement into account.

Enabling raytracing support on the other hand would improve 3D lightning effects alot. Given there is little environment in the BOTF combat scenes, it however won't differ much on what can be achieved with traditional lightning effects.

Would be awesome when in addition AI would add asteroid and nebula environments to certain sectors, but even when some AI would be capable of it, it would need access to the relevant information like e.g. the sector position and stellar objects.

Worse however is that BOTF already projects all rendered graphics into 2D space before passing it to the rendering device. Therefore no lightning effects can be post-applied at the device driver level.

What AI could be used for is to up-sample images or textures, or possibly help generate new ones. I've seen amazing samples where not only they increase resolution, but add matching detail when changing image aspect. But like Tethys mentioned, it doesn't know on the WDF interface definitions used by BOTF. And while GUI interfaces is the main visible graphics part with BOTF, AI doesn't understand how to properly arrange the GUI by usage.

Leaves race images, ship and building images that might be improved (or replaced) with AI. A few selected interface components possibly.
Best use I guess is to generate images for new structures where we don't have any yet.

Regarding 3D graphics on the other hand we already have awesome 3D models ready, that got down-sampled to make them work with the old BOTF engine. Texture detail is not what they miss. The old engine however is pretty limited on texture size and color palettes here.

Yes I hope this becomes a priority so we can have good detailed high resolution combat scences soon.

Not all is lost, we just need to properly replace the BOTF rendering engine and raytracing and all is possible.
Lots of analysis I did in the past, with promising results. But it'll be alot of work though and AI will be of no help there.

[Tethys]

This topic had me curious and I did a little Googling. There exists a program called DeepDi which is an AI based binary disassembler. I assume it would work on trek.exe as well and possibly expedite disassembly processes. I have no idea how it works and I don't plan to investigate much further due to lack of time for learning the new disassembler. But perhaps someone is curious to have a look and maybe see if they can go deeper than a traditional disassembly. From what I can see, the program is free.

[Spocks-cuddly-tribble]

a program called DeepDi which is an AI based binary disassembler.

I think you misinterpret the context. Project goal is to be faster and less CPU/GPU demanding in the initial disassembling of unknown files (e.g. via simplified code context analysis). The main field of application is e.g. initial security analysis of unknown files for protection software. So real time analysis of all CPU/GPU/RAM tasks can be faster and resource saving, which might enable the max security realtime protection even for old/slow hardware (without unbearable system slowdown).

[LadyGray]

OK


Getting off topic from GRAPHICS but still about AI


How about getting the AI to be trained as a human player in a LAN party https://towardsdatascience.com/how-to-t ... 1489f3bc5c
We could do multiplayer with a custom trained AI as an alternative to the poor AI included with BOTF ?

[Flocke]

If BOTF was a simple snake program

I guess it will take a few decades until AI is capable enough to learn complex games like BOTF just from the rendered screen input...