I've Created a WC4 HD Video Pack

Can you enhance the AI algorithm by inserting keyframes? E.g say you were to photoshop some of the images to provide a higher-resolution source material. Can the algorithm blend those enhancements into the video? (from the Youtube clip one of the obvious things is the glyphs on the Emperor's throne) We have some high resolution still shots from Wing Commander 3's cutscenes that were released as part of promotional materials. We could also attempt to enhance some of the 3D models in some of the shots if the AI engine is capable of blending those improvements into the output.

Hi @Darkmage,

That's an interesting thought. I'm not sure it would make much difference, since the AI models don't read the stream in its optimised form - it literally renders out each frame (keyframe or otherwise) as a complete image and upscales from there.

However, it does compare surrounding frames for extra data. That's why I'm not sure. I'm happy to give it a try if you'd like to send some images over to experiment with :) I suspect, though, that unless we could provide every fourth or fifth frame, it won't give it enough new data to work with.
 
MV5BMmM5M2NkNzEtMTk3Ny00ZWFjLTllOGEtNGVjNDUxMmUxZWY4XkEyXkFqcGdeQXVyMTAxODYyODI@._V1_.jpg

This is one of the promo shots. I can only assume the entire game was originally rendered in this resolution and then compressed down.
MV5BMzYyOGQxMTktNjc1ZS00NGI0LWIyZTgtMzM1Mzk3YTdkYTQzXkEyXkFqcGdeQXVyMTAxODYyODI@._V1_.jpg

MV5BZjhhZTJmYWMtNmIyMC00M2I4LThkNzQtOWI3NWZkNWQxZWVhXkEyXkFqcGdeQXVyMTAxODYyODI@._V1_.jpg

MV5BZmY5YTM0OTQtODE3YS00NzRkLTllYTYtNTk4OGRlYWE3MGVmXkEyXkFqcGdeQXVyMTAxODYyODI@._V1_.jpg

I know there's more pictures out there. These are from IMDB. But I've seen others. There's a promo catalog that's on the Privateer CD that has more. Alternatively, if someone has the original 3D assets, I have a working SGI O2 machine here and a copy of PowerAnimator... I can also get access to a second SGI O2 machine.
 
Hi @Darkmage,

I'll have a tinker with the Concordia crash scene and see if it helps the AI model at all. I suspect it won't, but I'll have a play around and get back to you :)
 
So, I had a look at that WC3 intro. Initially, looking at the credits, and the carrier jumping in, I was like "meh".

Then the Kilrathi showed up on screen. My jaw dropped upon seeing their hair actually looking like hair. Given the source material - I am utterly blown away.
 
So, I had a look at that WC3 intro. Initially, looking at the credits, and the carrier jumping in, I was like "meh".

Then the Kilrathi showed up on screen. My jaw dropped upon seeing their hair actually looking like hair. Given the source material - I am utterly blown away.

Thank you :)

The credits could definitely be improved. This was something of a quick experiment mainly to test if I could improve on the eyes and teeth from the previous Rollins example. I only spent six hours on this 🤣
 
These high res videos are so awesome; I've started to replay WC4 and WCP after so many years just to enhance my player experience. There must be a way to patch high res videos with Priv2 and WC3. I can't say enough good things about ODVS's work. Didn't think that these games could have so much new life breathed into them after so many years.
 
These high res videos are so awesome; I've started to replay WC4 and WCP after so many years just to enhance my player experience. There must be a way to patch high res videos with Priv2 and WC3. I can't say enough good things about ODVS's work. Didn't think that these games could have so much new life breathed into them after so many years.

Aw shucks, you'll make me blush 😊

Thank you
 
I'll see what else I can dig up in terms of game footage. I also have access to the 3DO copy of the game. Might be able to get another source of video to cross reference with the PSX clips.
 
I'll see what else I can dig up in terms of game footage. I also have access to the 3DO copy of the game. Might be able to get another source of video to cross reference with the PSX clips.

Oh thank you, it would be great to have some 3DO footage just to experiment with, to see what difference it makes. Do you think you could rip the same intro video and convert it to something readable? An uncompressed AVI would be great (though likely huge), but if that turns out to be too big to share, anything with a high bitrate could do - a high bitrate h264 MP4 or ProRes MOV for example 🙂
 
Oh thank you, it would be great to have some 3DO footage just to experiment with, to see what difference it makes. Do you think you could rip the same intro video and convert it to something readable? An uncompressed AVI would be great (though likely huge), but if that turns out to be too big to share, anything with a high bitrate could do - a high bitrate h264 MP4 or ProRes MOV for example 🙂
I've got a 40/20 internet link here. Getting the file out will be fine, the harder thing is getting a readable file. I've already extracted the .BUN from the 3DO cd but I have no way to unpack that and get out the individual files. if HCL could jump in, we could get this solved pretty quickly. He has more knowledge/utils for getting these files into readable form. Alternatively I can do an S-Video capture off my 3DO and put a file up, but I'm worried that will add unnecessary artifacts/compression. BTW if you checkout this video: https://www.wcnews.com/holovids/trailers_wing3.shtml it's higher quality than the ingame WC3 footage, and adds more scenes, particularly the shot of Confed HQ that didn't quite upscale as nicely...
 
Last edited:
BTW if you checkout this video: https://www.wcnews.com/holovids/trailers_wing3.shtml it's higher quality than the ingame WC3 footage, and adds more scenes, particularly the shot of Confed HQ that didn't quite upscale as nicely...

Unfortunately, I've found there are a few no-nos when using the deep learning routines for upscaling footage. Some of them are:
  1. Anything captured from VHS (and most analogue tape-rather-than-film sources) are a no go. It's not really the noise and visual hiss interference introduced by the medium that's the problem - such sources tend to lack defined edges. The AI routines have learned to preserve blurring when upscaling (this helps maintain things like focus blur/motion blur/bokeh etc.). So if there's a soft line of delineation between, say, someone's cheek and the backdrop behind them, that soft line will be replicated in the upscale - and your results won't be much improved over traditional bilinear/scalar upscaling methods (if at all).

    Annoyingly, most attempts to digitally sharpen footage (unsharp masking routines and the like) introduce their own artefacts to the image by the very nature of how they work (ghost effects around edges etc.), which then get preserved too. I've been getting a lot of that in my various attempts to clean up the WCP vids before upscaling them. This is one of the reasons the WC4 upscales are so much more successful - 35mm film FTW!

  2. For similar reasons to above, you get pretty unimpressive results if you're trying to improve footage that's already been upscaled by traditional means.

    I've tried downscaling content that's been traditionally upscaled (taking the 1080p stream from a Blu-Ray release of a 90s TV show like Red Dwarf, for example) back to Standard Definition then running it through the AI models - but the damage has already been done. Traditional upscaling softened all the defined edges, downscaling that footage didn't resharpen the edges, the AI model maintains the damage.

  3. Due to the above you tend to get considerably better results from a grubby original-resolution capture than you do from something that has been "cleaned up" for Blu-Ray or online streaming. The AI models will themselves clean up a lot of things like colour banding, noise etc. - but if you don't have defined edges, you're not going to get a high quality image in the upscale.

    So although it's counter-intuitive, if the original source material isn't HD you get better results from, say, an original 90s DVD release than a modern Blu-Ray release.
Double unfortunately, I think every video on that page contravenes at least one of the issues above :(

I'd definitely be interested in getting my hands on some 3DO videos if anyone can manage to rip them. Any improvement over the PSX footage in resolution (or especially frame rate) could have a major impact on end results 🙂
 
Oh thank you, it would be great to have some 3DO footage just to experiment with, to see what difference it makes. Do you think you could rip the same intro video and convert it to something readable? An uncompressed AVI would be great (though likely huge), but if that turns out to be too big to share, anything with a high bitrate could do - a high bitrate h264 MP4 or ProRes MOV for example 🙂

The issue with the 3DO version right now is that we don't have a decompressor for it. Having the videos on hand would be useful for experimenting, but WC3 doesn't use a standard format and some of the existing documentation on the format is kind of misleading since it was later revealed that there's two different formats: one used in the 3DO demo that's closer to the PC format, and one in the final game. The 3DO used some kind of hardware video playback that the video files call. @HCl worked on a player for them, had promising results, but he never released a completed version.
 
Back
Top