Duke4.net Forums: icecoldduke - Viewing Profile - Duke4.net Forums

Jump to content

Hide message Show message
Welcome to the Duke4.net Forums!

Register an account now to get access to all board features. After you've registered and logged in, you'll be able to create topics, post replies, send and receive private messages, disable the viewing of ads and more!

Reputation: 1329 My posts turn threads into gold
Active Posts:
2,129 (0.61 per day)
Most Active In:
Everything EDuke32 (723 posts)
21-December 09
Profile Views:
Last Active:
User is offline Jun 27 2019 09:44 AM

My Information

Age Unknown
Birthday Unknown
Not Telling Not Telling

Contact Information


Posts I've Made

  1. In Topic: Ion Fury

    11 June 2019 - 08:09 AM

    I just read about the lawsuit. I personally think it's bullshit and I wish Richard, Evan, and the entire team the best of luck. Unfortunately these guys can keep you in legal hell for the foreseeable future, which is beyond disgusting. This game needs to get released, I'm at a loss on words. Anything I can do guys let me know.
  2. In Topic: PolymerRTX

    10 March 2019 - 11:12 PM

    What I want to use is something like ESARGAN or something similar to upscale dynamically created light maps that are created async at runtime.

    ESARGAN, TopazAI etc are best at doing real life image upscaling, because there AI models are trained that way. Yes you can get 30 to 60% good quality ups along by shoving Duke sprites or lightmaps through public ally released ESARGAN models but they donít deal with stylised art that weíll. Iím doing research to see if I can fix that problem for the above purpose.
  3. In Topic: PolymerRTX

    10 March 2019 - 12:37 PM

    So just a little bit of an update, I'm continuing to try and train the neural network to deal with some of the graphics errors I've found when processing Duke3D sprites. I'm getting close to releasing something for you guys to run on your end, to test and get feedback. For additional neural network training, I've been grabbing a lot of old game data, manually creating some hd textures, and training the network to create my shitty hd version of various things :). It's going well. I know you guys want to see want to see more Duke bits, and hopefully I'll have more to show after I train the neural net a bit more. In the mean time, here is some progress. I can take Iron Helix, that is a Myst style adventure game, were everything is rendered as pre-rendered movies approx. 162x122), and 4x it's size while adding detail. I'm still getting some consistent edge case fails, across all games I've thrown at this(space and planets don't seem to come through properly for example), anyway here are some screens.

    As I side note, I must have played the intro cinematic 1000's of times as a kid, and I never realized there was a smiley face easter egg on the radar screen, that represented the planet the ship was about to destroy. I've found quite a lot of interesting things running various games through this tool.

    Posted Image

    Posted Image

    Posted Image

    Posted Image

  4. In Topic: LIZTROOP !

    28 January 2019 - 11:04 AM

    It's cool I guess, but I'm not sure what I'm looking at. Did you just re-art the liztroops and hook up there ai? As others have said, those projectiles are really off.
  5. In Topic: PolymerRTX

    22 January 2019 - 12:24 AM

    View PostMusicallyInspired, on 21 January 2019 - 10:38 PM, said:

    They're upscaled by the engine during run-time. He doesn't have a library of upscaled sprites....at least I don't think he does. If he did why would he even post this as "progress?" His renderer is upscaling the low-res vanilla sprites on the fly.

    To be clear, this isn't real time. So yes I have a huge collection of AI upscaled tiles sitting on my hard drive. So why is this "progress"? I actually started doing this because I wanted to figure out how the core part of DLSS could work(since there really isn't a lot of documentation on how it works). There are tools out there now, that you can use to do AI upscaling, but I guess the cool part is, I figured it out myself :), which isn't something I should have called "progress", so sorry, but I'm excited that I figured it out(yay me :)) Also this is my first time using a neural network that generated some pretty cool results, that executes on the RTX hardware.

    But no I'm not releasing a sprite dump because I'm pretty sure that's illegal :).

    Let me catch everyone up on what I've been up too. This past month has mostly been about, how can I use RTX as a development acceleration tool. So a couple things came to mind for testing, image upscaling(see above) and lightmapping. Both bits have been done for years, but not with hardware acceleration. So I made a Return to Castle Wolfenstein lightmapper that can fully light a map, that has 571 lights, in about 6.5 seconds, at 4k. The lighting results in this scene don't match the original(I'm doing some testing where I needed to exaggerate some lights, I post some more stuff when I get back to the U.K.). The trashed looking triangles are because I'm not handling Quake's tessellated surfaces properly.
    Posted Image

    So how does all of this apply to PolymerRTX? I'm trying to see if I can expand the renderer out, so devs might require a 2080 to make a map for PolymerRTX, but you can use a 2060 to play. Essentially light baking using a properly trained neural network. I don't know if this can work, I'm essentially in uncharted territory. So basically the above two projects are basically me trying to learn two very complicated concepts, and coming with a new type of rendering idea, and it might not work out, but what I learn here can translate into something more tangible.

    I have to head out to my flight, but I'll try to catch you guys up on the idea when I can, I'm sorry I've just been traveling all over the place as of late.

All copyrights and trademarks are property of their respective owners. Instead of reading this text, you could be playing Ion Fury! ;) © 2019 Voidpoint, LLC

Enter your sign in name and password

Sign in options