Star Trek: Deep Space Nine Documentary in the Works

Yeah, Kes was trying to be a "Troi" type character that didn't really work. Chokatay was just an exercise in bringing in some politically correct Native American character as that really can't been done yet in a Trek crew...worked okay sometimes but too cliche most of the time.

One thing I thought they should have done from the beginning was instead of casting Robert Duncan McNeill as Tom Paris was to cast him in the role he played in the TNG episode he was in (Nick Locarno) only after his whole thing there - there was a story in there, how he dealt with getting on with Star Fleet...and it would have first with his character in Voyager fairly easily if they wrote it that way.

 

That would have required Paramount to pay the author of the TNG episode that introduced Nick Locarno a large heap of money. That wasn't going to happen.

 

Actually, the episode Trials And Tribble-ations was included in HD in the original series blu-ray set as a bonus feature.

 

Mulgrew wasn't as good at selling technobabble.

 

(Which was unfortunate, since Voyager was absolutely saturated with it.)

 

Badly upscaled. Not real HD like we see in the remastered "Birthright".

 

The Sisko rests with the Prophets.

 

I watched Shatner's documentary "The Captains" and after watching the one with Avery (Sisko/DS9), I can kinda understand why he didn't (or wouldn't) appear in the documentary the OP made.

 

I always saw her as the "Manager", and as such, no need for geekspeak. One of my favorite radio station managers was non-tech-y as a...non-tech-y thing. But, he could read people, make a good snap decision, and interpret what people were telling him.

 

https://trekmovie.com/2019/04/07/st...d-to-screen-in-theaters-for-one-night-in-may/

The docu will have DS9 footage in HD. Looking forward to that.
I would definitely rewatch both DS9 and Voyager if they get the TOS and TNG treatment

 

Yes, it's a shame that Enterprise easily got a remaster because that show was already broadcast in widescreen. Sure DS9 and VOY would be more expensive to transfer to HD plus having to update the effects like TNG. Still hope they get to do it though...

 

It seems like Voyager and especially DS9 is picking up new fans all the time. Star Trek is popular again and it's easily available on netflix.
Still, a long way to go before they work on those shows again. To many episodes. To expensive.

 

A fan played around with using AI to upscale DS9 off DVD. Doesn't look as good as a full HD remaster, but I'm betting if you did this using a fresh digital transfer of the original analog masters, it would look a hell of a lot better than the current, crappy digital masters available on the DVDs and from streaming services like Netflix.

And it would cost very little to do compared to a full HD remaster.

 

Sound is slightly better on dvd but I'm glad it's streaming so I don't have to deal with NTSC and PAL.

Stargate SG1 isn't streaming so I bought the dvd box in PAL and the 4% spedup sucks

 

Here's a sample of the machine learning driven "HD" remaster of DS9.

 

Somebody else just boosted the sharpness and contrast - I always thought DS9 looked washed out and fuzzy - and produced a much more pleasing result than the standard remaster. I think if you combine that with the machine learning uprezzing you could have a decent picture even on fairly large screens. And it could be done for a fraction of what a real remaster would cost:

 

I couldn't play that video, but found this one that worked:

DS9 Remastered With Machine Learning - Comparison Video (480p to 1080p)

To me, there's not much difference between them. I don't know how their machine learning algorithm works, but frankly I was able to achieve better results when I remastered the series for myself (without no AI), back some years ago.

I had to get rid of interlacing and blended frames, reduce noise, boost contrast/gamma and add a little sharpness. It took ages, but the results look pretty darn good.

 

I could see the difference in the first video, but the second video was more noticeable.

Darryl

 

I meant to say I could not see the difference in the first video.

Darryl

 

I notice a big difference on both videos, but I've got a 34" computer monitor to view them on.

Would love to see the videos processed as indicated - get rid of interlacing and blended frames, noise reduction, boost contrast/gamma, sharpen - and then fed to machine learning to upres. I'm betting the combined results would be quite dramatic.

 

The machine-learning upgrade is amazing. It tightens up fuzziness, seemingly without creating any edge haloes. I’m sure every company that does upscaling will be trying to develop this.

Of course if this is the next breakthrough in upscaling, it means all the shows that have already invested heavily in being upscaled using the best available techniques, like Farscape and Red Dwarf, may eventually need to be bought again if they’re upgraded and if you’re a perfectionist.

 

Hmmm, are we watching the same video? Because I can only see a very minimal difference between the two versions. Nothing I would characterize as 'amazing'.

Also, I'd hate to think how long it takes to process a video if a neural net is involved in learning, because it would have to be processed perhaps many thousands of times in order for the NN to converge on an optimal solution and the same process may need to be repeated for each different video (episode).

If those assumptions are true, then I doubt this will ever become a speedy, general purpose solution alternative to conventional computational techniques.

 

How big is your monitor? On my 34" screen the improvement is immediately apparent, even in that little split screen, which is only half the resolution of a full 1080p image.

I'd like to see them switch back and forth between the two images on a full 1080p field.

That's likely not how it works, at all, since they don't have access to any 1080p DS9 footage to train the model with. They probably trained the model using 480p and 1080p copies of the same video - not DS9, but likely a downscaled Blu-ray of a film or TV show - and trained the model to make the tweaks necessary to make the 480p video look as much like the 1080p original as it could. It therefore "learned" what changes worked and what changes didn't, and how to make a 480p image look like the 1080p original.

They then unleashed it on 480p DS9 video, and it used the same tricks to uprez it to 1080p.

 

Physical size doesn't really matter, what's important is the native resolution of the monitor. I have a 24" 1920x1200 pixel 16:10 monitor, which is slightly larger than HD, the max resolution of the clip. So it's a 1:1 clip/pixel correspondence.

They do actually, at least a little bit. When they did the TNG remaster, there were some scenes on DS9 that were a part of it, so they could have used that (not sure if it was sufficient).

Possibly, but I'm not sure how well that would work. There are all kinds of different encodings for different movies and release versions. These encoding differences might contribute to what the NN learns for that particular release and may not translate well to a TV show with different attributes.

My strategy would be to choose the closest source material for which there is an exact match between original DVD release and remastered bluray release, such as TNG (which also includes scenes from DS9).

But these are all guesses since we don't know what methodology was actually used.

 

It absolutely matters. On a small monitor it can be difficult to impossible to make out any resolution differences between two different images, especially when the image is scaled to only take up less than half the monitor to begin with. Those differences can become obvious on a larger monitor. It's pretty difficult for example to make out the difference between 720p and 1080p content on a 1080p 27" monitor. On a 56" monitor of the same resolution, though...

Train the model on lots of different clips from lots of different shows. Then it'll learn what works and what doesn't across all of them. Heck, it might even learn how to detect differences between source pictures and switch to different techniques as it sees fit.