so where's all the Topaz Video Enhance AI discussion?

[pm-s]

Upscaling is just a gimmic.

[Selur]

Yeah, main problem is to filter a source to remove problems like ringing, blocking, etc. while keeping enough details which later could be enhanced during upscaling otherwise you get that 'plastic look'.

[lordsmurf]

Upscaling is just a gimmic.

Mostly, yes.

The only real exception is where you have mixed-content projects (almost exclusively documentaries). Those are valid.

But the vast majority of people doing upscale is for pointless nonsense.
- Too many "restore" bottom-quality Youtube videos.
- Many more try to upscale VHS -- and from a bad capture that had no TBC nor a quality VCR. It's crap quality "blown up".
- Few actually upscale retail DVD sources, aside from scenes for editing sources (again, mixed content).
- Recordings of video games, mostly from consoles that really weren't even base SD quality.

Far too many people also confuse upscale with "crank the sharpen knob to 11". You can butcher sharpness in SD already, HD not needed.

All of this pointlessly bad upscaling get amplified when the upscaler sucks (ie, Topaz).

FYI, Youtube is too often used as a yardstick for what to do. For example, "Youtube insists on 1080p uploads". Well, so what? Youtube has never been the yardstick for quality. They can (and do) whatever dumb thing they want, in terms of encoding. Long-term, it's always wasted effort. They change ingest specs as often as some folks change their underwear (not literally, I hope <eww>, but you get the idea). So upscaling to appease Youtube is equally pointless.

Upscale is never straight-forward. It doesn't simply make video better, but merely different. New artifacts emerge, and the net result can be worse.

I've been upscaling images (still and video) since fractal methods came out in the 90s. And not much has changed over the years. It's the same basic artifacts that you're fighting, and that's unlikely to ever change. The only reason that upscale gets "better" is merely because the sources get better. Not some magical BS like "AI".

[johns0]

The only way ai could properly upscale videos if it was able to imagine what the video could look like and basically redo the video so it was actually 1920x1080 for example but there is no such thing in any ai that can do that now.As lordsmurf has said it's just all sharpening that people look at and think it's better.

[pm-s]

is nnedi3_rpow2 considered upscaler?

[Selur]

nnedi3_rpow2 is for enlarging images by powers of 2.

source: http://avisynth.nl/index.php/Nnedi3/nnedi3_rpow2

[DruidCtba]

Guys I need a lot of help here with the Hybrid :/.

It turns out that in the Topaz community, I received a file that I had sent in 480i and someone improved it using some parameters of the QTGMC filter to 480p, but when using the same parameters (my friend's machine was a MAC running Hybrid) it gave a CRASH error at a certain point.

After discovering that the files that I had been testing with some Topaz models, in reality and although MPCHC and MediaInfo claim to be progressive, they are not, and how do I know? I used this script here: http://www.aktau.be/2013/09/22/detecting-interlaced-video-with-ffmpeg/ and verified that really, only the file my friend sent me truly reflects being progressive, generated by Hybrid :/.

And looking closely, I can see that the files I generated without using the parameter "overwrite input scan type to: bottom field first", the comb/alias effect of the interlaced files is visible.

The fact is that when I use the parameter "overwrite input scan type to: bottom field first" I get a CRASH error on Hybrid and it doesn't convert the file to progressive, it doesn't actually generate any file at all, and I don't know how to solve this, so I'm counting on the masters here to help me unravel the reason for this error and how to solve it on my Windows 10 Pro 21H1.

Here are my Hybrid settings, where am I going wrong?

https://drive.google.com/drive/folders/1L2gHLwIlWnhl2UuMhMQv7XpXsSPYuZgK?usp=sharing

I've even uninstalled and reinstalled the Hybrid, but it keeps giving the CRASH error .

Thanks in advance for any help .

Att.

Druid®.

[DruidCtba]

Upscaling is just a gimmic.

Mostly, yes.

The only real exception is where you have mixed-content projects (almost exclusively documentaries). Those are valid.

But the vast majority of people doing upscale is for pointless nonsense.
- Too many "restore" bottom-quality Youtube videos.
- Many more try to upscale VHS -- and from a bad capture that had no TBC nor a quality VCR. It's crap quality "blown up".
- Few actually upscale retail DVD sources, aside from scenes for editing sources (again, mixed content).
- Recordings of video games, mostly from consoles that really weren't even base SD quality.

Far too many people also confuse upscale with "crank the sharpen knob to 11". You can butcher sharpness in SD already, HD not needed.

All of this pointlessly bad upscaling get amplified when the upscaler sucks (ie, Topaz).

FYI, Youtube is too often used as a yardstick for what to do. For example, "Youtube insists on 1080p uploads". Well, so what? Youtube has never been the yardstick for quality. They can (and do) whatever dumb thing they want, in terms of encoding. Long-term, it's always wasted effort. They change ingest specs as often as some folks change their underwear (not literally, I hope <eww>, but you get the idea). So upscaling to appease Youtube is equally pointless.

Upscale is never straight-forward. It doesn't simply make video better, but merely different. New artifacts emerge, and the net result can be worse.

I've been upscaling images (still and video) since fractal methods came out in the 90s. And not much has changed over the years. It's the same basic artifacts that you're fighting, and that's unlikely to ever change. The only reason that upscale gets "better" is merely because the sources get better. Not some magical BS like "AI".

Look my friend lordsmurf, I agree with you and respect your opinion for being a professional for years in this area, editing videos, since I am an IT Systems Analyst, who knows very little about this universe of video and photo editing, I just use photoshop basics for editing photos and some commands in DaVinci Resolve to tidy up some videos for my enjoyment, not like you being a pro, but let me give you the opinion of an end user, not a pro OK.

From my point of view, I just want to have a better definition of my Show DVD's, on my TV Samsung TU8000 4K 50K, which unfortunately didn't come out of HD/FHD/UHD, and I confess that I'm excited with some results, both using Avisynth like Topaz.

That's important to me, because it's a personal hobby, as it must be for many who like old series that didn't come out in HD/FHD/UHD either, and that's what makes all the difference.

Of course I know there is no magic, and it seems that everything now revolves around AI, and they advertise everything, but the reality is that AI still comes at a very slow pace, at least as far as the area of image and videos and their improvements, there are a lot of lies about AI in these two areas, what I've seen is that these so-called AI softwares just INVENT where they don't exist or existed and it was cut so for me, who work with systems, I know AI is learning above of what existed, it requires a real model, to be able to create and/or invent what no longer exists in a compressed source and as you very well report in a "committed" source on Youtube.

The fact is that I have achieved a final product that I like and it is much better than taking a DVD and using Upscaling my Samsung TV to see a show than a 1080p or 2160p generated by an avisynth script like friend selur modeled or a Topaz of life.

I would love to encourage you to do the same as our friend selur, since you as a professional know what you can and cannot improve using properly using avisynth or vapouravisynth scripts, if it is of interest to you of course, we all appreciate it, and any help is welcome, so it is possible to unmask companies that use the darling of the moment called IA, which actually only masks the use of avisynth, vapouravisynth, ffmpeg, or other big name open source programs that are used behind the name IA.

Att.

Druid®.

[aedipuss]

druid

can you make a 10 dollar bill into a 100? no? well neither can any software make low rez vid into high rez. the necessary data/quality just isn't there. period.

[DruidCtba]

druid

can you make a 10 dollar bill into a 100? no? well neither can any software make low rez vid into high rez. the necessary data/quality just isn't there. period.

Friend and I said something different than what you're saying, go back there and read my post and you'll see that I said AI is walking slow when it comes to image enhancement and videos, I think you're another one of those with blood in your eyes when you hear someone here talking about Topaz, and look that I criticized the product and you didn't even notice such is your cognitive dissonance, mercy, deliver me sir from these types of people .

Att.

Druid®.

[Selur]

Guys I need a lot of help here with the Hybrid :/.
..
Here are my Hybrid settings, where am I going wrong?

https://drive.google.com/drive/folders/1L2gHLwIlWnhl2UuMhMQv7XpXsSPYuZgK?usp=sharing

Each Hybrid user log starts with:

This log is only intended for user information. It should not be part of a bug/problem report!!
Hybrid: homepage
Forum: public forum
Regarding problems please read: needed infos

So if you want help: Follow the 'needed infos', read it, understand it, follow the instructions.
Either properly post with the needed infos in the 'public forum' or post into the Hybrid thread here in the forum.
If you do not share the necessary details (= description of what you want to do, what you did and a debug output of it), there is no helping you.

Cu Selur

[pm-s]

I like how upscaling is just a small portion of what people consider as upscaling, and how that "upscaled look" comes from smart sharpening rather than resizing algorithms.

[DruidCtba]

Guys I need a lot of help here with the Hybrid :/.
..
Here are my Hybrid settings, where am I going wrong?

https://drive.google.com/drive/folders/1L2gHLwIlWnhl2UuMhMQv7XpXsSPYuZgK?usp=sharing

Each Hybrid user log starts with:

This log is only intended for user information. It should not be part of a bug/problem report!!
Hybrid: homepage
Forum: public forum
Regarding problems please read: needed infos

So if you want help: Follow the 'needed infos', read it, understand it, follow the instructions.
Either properly post with the needed infos in the 'public forum' or post into the Hybrid thread here in the forum.
If you do not share the necessary details (= description of what you want to do, what you did and a debug output of it), there is no helping you.

Cu Selur

OK, thank you my friend selur, I will do as you are directing me, thank you so much for taking your precious time in answering me.

Att.

Druid®.

[DruidCtba]

I like how upscaling is just a small portion of what people consider as upscaling, and how that "upscaled look" comes from smart sharpening rather than resizing algorithms.

Look, I have two very clear points of view about Upscaling, especially when it comes to Upscaling with AI.

1. From a professional point of view and I do not disagree in almost anything with the professionals who respond here, who are knowledgeable in depth on the subject, I have enormous respect for everyone who somehow elucidates these points, but I do not look favorably on those who just come here to criticize an A or B product without giving solutions or even explaining at a more learning or noobie level, thus making people understand in a more colloquial language, because in my IT area I have to do this, all the time, rather than using computations or technical terms that will only confuse more than clarify such points.

2. From the amateur point of view, as I explained above, for many what you are seeing as a positive result of better entertainment is worth it, which satisfies you most as a final product, and that's enough, be it an avisynth implementation or even a Topaz, it's like someone saying that dark chocolate is good for being natural, when most will certainly prefer semisweet or even sweet, what matters is what gives you the most pleasure and not how to dictate to you what you should give you pleasure, and I say this in amateur terms, anyway...

Att.

Druid®.

[lollo]

I like how upscaling is just a small portion of what people consider as upscaling, and how that "upscaled look" comes from smart sharpening rather than resizing algorithms.

https://forum.doom9.org/showthread.php?p=1810955#post1810955

[pm-s]

druid

can you make a 10 dollar bill into a 100? no? well neither can any software make low rez vid into high rez. the necessary data/quality just isn't there. period.

Friend and I said something different than what you're saying, go back there and read my post and you'll see that I said AI is walking slow when it comes to image enhancement and videos, I think you're another one of those with blood in your eyes when you hear someone here talking about Topaz, and look that I criticized the product and you didn't even notice such is your cognitive dissonance, mercy, deliver me sir from these types of people .

Att.

Druid®.

While not being awful, TVEAI does deserve criticism for being paid program that gives you no manual options

[Selur]

...that gives you no manual options

Please explain in detail how they could do that while using ML based filtering?

I mean ML simplified works like this:
- Train model, by showing it lots of broken and corrected content.
- During filtering you apply that trained model on new content and hope for the best.
Unlike normal filtering through 'simple' algorithms there is no way to have tons of parameters to change the output. And last time I checked they do offer some setting which basically lets the user select some different models,...

-> From my point of view your your argument seems baseless.

Cu Selur

[pm-s]

...that gives you no manual options

Please explain in detail how they could do that while using ML based filtering?

I mean ML simplified works like this:
- Train model, by showing it lots of broken and corrected content.
- During filtering you apply that trained model on new content and hope for the best.
Unlike normal filtering through 'simple' algorithms there is no way to have tons of parameters to change the output. And last time I checked they do offer some setting which basically lets the user select some different models,...

-> From my point of view your your argument seems baseless.

Cu Selur

No, I was talking about preprocessing. It denoises perfectly fine looking footage, giving it that plastic look. And there is no way you can disable denoising

[Selur]

No I was talking about preprocessing. It denoises perfectly fine looking footage, giving it that plastic look. And there is no way you can disable it

I doubt there is additional pre-processing, I'm quite sure that simply is what applying their trained models does.
Same when using BSRGAN and other ml based models for upscaling.
Topaz is basically a gui for something similar to TecoGAN a temporally coherent GAN approach.
Their marketing team does a good job creating a hype by presenting good results with content which works fine with content that is similar to what the model was trained with and not mentioning the problems you might get depending on what you apply that model on.

Cu Selur

[pm-s]

You've got a point

[pm-s]

The thing is, most people with that software want to restore something way worse than 720 or sharp 480px video with very little compression applied on. That's where problems start kicking in.

[DruidCtba]

druid

can you make a 10 dollar bill into a 100? no? well neither can any software make low rez vid into high rez. the necessary data/quality just isn't there. period.

Friend and I said something different than what you're saying, go back there and read my post and you'll see that I said AI is walking slow when it comes to image enhancement and videos, I think you're another one of those with blood in your eyes when you hear someone here talking about Topaz, and look that I criticized the product and you didn't even notice such is your cognitive dissonance, mercy, deliver me sir from these types of people .

Att.

Druid®.

While not being awful, TVEAI does deserve criticism for being paid program that gives you no manual options

Yes friend I agree that his price is too high for what he does, but now it seems things are getting a little better, they have released a new model called Proteus, VEAI 2.3.0, where you can change 6 parameters manually.

Att.

Druid®.

[DruidCtba]

I like how upscaling is just a small portion of what people consider as upscaling, and how that "upscaled look" comes from smart sharpening rather than resizing algorithms.

https://forum.doom9.org/showthread.php?p=1810955#post1810955

Amazing restoration work, too bad he didn't want to show how he did the same, it only served to caress his ego. I believe that being a professional and fearing the competition, about his model, I don't take his credit for it, after all he must have spent a lot of time to get this result, so it only serves to caress his ego on the forum, in fact it has become a constant in these forums unfortunately.

Att.

Druid®.

[lollo]

too bad he didn't want to show how he did the same

You must be joking, videofred published all his scripts several times, and were used as the starting point for video restoration by many of us...

[DruidCtba]

No I was talking about preprocessing. It denoises perfectly fine looking footage, giving it that plastic look. And there is no way you can disable it

I doubt there is additional pre-processing, I'm quite sure that simply is what applying their trained models does.
Same when using BSRGAN and other ml based models for upscaling.
Topaz is basically a gui for something similar to TecoGAN a temporally coherent GAN approach.
Their marketing team does a good job creating a hype by presenting good results with content which works fine with content that is similar to what the model was trained with and not mentioning the problems you might get depending on what you apply that model on.

Cu Selur

Yes friend Selur, you are correct on this point, the principle of an AI is to train with original models and their compressed productions (VHS, LD, DVD, TV and so on), and the greater the number of original models the better an AI works, what happens as you said is that they would need, for example, an astronomical range of original content, things they don't have, to be able to greatly improve their models, ie they would have to, for example, for movies, series and shows to have access to the original tapes of the studios to promote training with the results obtained in VHS, LD, DVD, TV, products launched on the market, and this has a very high cost that they will never afford, so what happens here is that the product will never be enough good for the professional world, but for amateurs and hobbyists it will be, at least from my point of view.

Att.

Druid®.

P.S. Look at the work presented above using avisynth and/or ffmpeg, the parameters used by the professional who made such restoration achieved, I believe, a better result than perhaps would be achieved today by VEAI, but this same result may not be applied to your video in particular, whatever you are wanting to restore, it's all based on trial and error, and that takes time and that's why I said that I don't take credit for this professional in not revealing the parameters he used.

Att.

Druid®.

[DruidCtba]

too bad he didn't want to show how he did the same

You must be joking, videofred published all his scripts several times, and were used as the starting point for video restoration by many of us...

Friend, I confess that I don't know about this, after all I only read his post and some ahead and he himself said that he would not reveal how he reached such a result, but that they could speculate on the models he made available on the forum, don't get me wrong, as I said, I understand perfectly his position, but at this point his topic shows what a professional quality he is, excellent for this job, but he only served to stroke his ego on the forum, but that's just my point of view .

Att.

Druid®.

[pm-s]

No I was talking about preprocessing. It denoises perfectly fine looking footage, giving it that plastic look. And there is no way you can disable it

I doubt there is additional pre-processing, I'm quite sure that simply is what applying their trained models does.
Same when using BSRGAN and other ml based models for upscaling.
Topaz is basically a gui for something similar to TecoGAN a temporally coherent GAN approach.
Their marketing team does a good job creating a hype by presenting good results with content which works fine with content that is similar to what the model was trained with and not mentioning the problems you might get depending on what you apply that model on.

Cu Selur

Yes friend Selur, you are correct on this point, the principle of an AI is to train with original models and their compressed productions (VHS, LD, DVD, TV and so on), and the greater the number of original models the better an AI works, what happens as you said is that they would need, for example, an astronomical range of original content, things they don't have, to be able to greatly improve their models, ie they would have to, for example, for movies, series and shows to have access to the original tapes of the studios to promote training with the results obtained in VHS, LD, DVD, TV, products launched on the market, and this has a very high cost that they will never afford, so what happens here is that the product will never be enough good for the professional world, but for amateurs and hobbyists it will be, at least from my point of view.

Att.

Druid®.

P.S. Look at the work presented above using avisynth and/or ffmpeg, the parameters used by the professional who made such restoration achieved, I believe, a better result than perhaps would be achieved today by VEAI, but this same result may not be applied to your video in particular, whatever you are wanting to restore, it's all based on trial and error, and that takes time and that's why I said that I don't take credit for this professional in not revealing the parameters he used.

Att.

Druid®.

It's not ai. It's machine learning.

[johns0]

Machine learning=ai,topaz aint.

[lordsmurf]

I just want to have a better definition of my Show DVD's,

Then buy newer HD copies. Because you're not making anything HD. You're just sharpening the video. You can do that without Topaz, and could have 20+ years ago. In fact, Topaz is slow, a good sharpener is fast.

old series that didn't come out in HD/FHD/UHD either

Appreciate it for what it is.
Restoration is reserved for making something better, and not perfect. Upscaling is most often a pointless exercise in seeking resolution perfection. Worse still, upscale has extreme tradeoffs, and is often NOT a net improvement to the content.

and it seems that everything now revolves around AI,

False. It's usually just a misused/raped buzz word. As you realize...

there are a lot of lies about AI

I would love to encourage you to do the same

No. For me, video is work. I don't want to always work. When video hobby time comes, I want to enjoy actually watching the (hopefully, and usually is) already-quality retail DVD video, not screw around with scripts to eek out a mere % sharpness gain. I'll only partake in hobby restoration projects when it needs actual restoration to be fully enjoyed (remove distractions, flaws, errors). Upscaling an otherwise already-quality retail DVD is insane, and I don't have that much time on my hands. I don't want to sit at a computer 24/7/365. If the content truly needs upscale, then the studio needs to go back to the source and re-scan.

Again, one of the few valid reason for upscale is mixed-source projects, aka documentaries. Not just resolution, but PAL/NTSC, interlaced, color/IRE matching, etc. Those are always great projects to work on, and you can have a real sense of pride when done.

too bad he didn't want to show how he did the same, it only served to caress his ego.

WTF? No. Fred shared, johnmeyer continued it and shared. Great guys, excellent film work.

I like how upscaling is just a small portion of what people consider as upscaling, and how that "upscaled look" comes from smart sharpening rather than resizing algorithms.

Yep. It's also very source dependent. Too many goobers try to upscale Youtube crap, rather than quality sources. The same goobers then wonder how somebody like Fred can do what he did, not understanding that his sources were actual film. To the goobers, it's all magic. They don't know what they don't know.

[pm-s]

So is it machine learning or not? Half of people are telling me it is, half are telling me it isn't