Topaz products

[taigi]

Are Topaz products any good?
Is there a possibility to make the same job using Avisynth scripts?

[lordsmurf]

Are Topaz products any good?

Mediocre at best.

Is there a possibility to make the same job using Avisynth scripts?

Yes. Avisynth is better.

[lollo]

Is there a possibility to make the same job using Avisynth scripts?

If you refer to Topaz Video Enhance AI you can't really compare to AviSynth. IMHO Topaz is "only" a resizer/sharpener with some noise reduction.

I experimented it in my restoration after the AviSynth scripts when I (rarely) want to upscale for watching the videos on my Smart TV.

I just replace the AviSynth code

nnedi3_rpow2(rfactor=2, nns=4, qual=2, cshift="Spline36Resize", fwidth=1440, fheight=1080)
addborders(240,0,240,0)

that is embedded in the full restoration script, with a Topaz VE AI pass with Gaia High Quality or Artemis High Quality model (without any noise/addgrain option active) at the end of the AviSynth processing.

Results are very similar, nnedi3_rpow2 combined with proper sharpening is excellent.

[taigi]

that is embedded in the full restoration script, with a Topaz VE AI pass with Gaia High Quality or Artemis High Quality model (without any noise/addgrain option active) at the end of the AviSynth processing.
Results are very similar, nnedi3_rpow2 combined with proper sharpening is excellent.

Sorry, just to clear this up, do you include Topaz VEAI together with the script you provided, or compare result of the script with the result of Topaz VEAI?

[lollo]

Some details:

workflow 1
restoration and upscale with AviSynth
- levels and colors correction before filtering
- denoising
- change color matrix
- upscale with nnedi3_rpow2
- sharpening
- levels correction after filtering
- compression with h264 (just to compare and upload the sample)

workflow 2
restoration with AviSynth (as per workflow 1)
- levels and colors correction before filtering
- denoising
- sharpening
- levels correction after filtering
- change color matrix
upscale with Topaz Video Enhance AI
-- avisynth output of previous steps is the input for Topaz

Some remarks
- Upscale is from 720x576 4:3 to 1920x1080 full HD
- I only used progressive videos; I ignore Topaz deinterlacing procedures, but I doubt it can be any better than QTGMC in AviSYnth
- The sharpening in the AviSynth step of workflow 2 should probably be slightly reduced to compensate sharpening occurring in Topaz
- It appears to me that some (additional) desoining is happening in Topaz
- The h264 compression done inside Topaz is out of control, you can only choose "Compression Factor" (bitrate allocation, I used middle setting)
- DAR parameter (4:3) has been added to the video input of Topaz with ffmpeg, otherwise Topaz will not respect width proportion
- For the attached sample I used Gaia High Quality model

Attached the results. The starting video capture is not that good, and the bitrate for workflow 1 is lower, but you can compare yourself



workflow1:
ufo_sII2a_spot_amtv_2_cut_v19_upscale_fullhd_reduced-muxed.mp4

workflow2:
ufo_sII2a_spot_amtv_2_cut_v19_reduced_cm_dar_1.88x_1920x1080_ghq-5.mp4

[pm-s]

- I only used progressive videos; I ignore Topaz deinterlacing procedures, but I doubt it can be any better than QTGMC in AviSYnth

https://youtu.be/LgZbxjF2AO4 its actually worse than QTGMC, messing up frames every now and then

[lollo]

Indeed

[Selur]

btw,... those that like Topaz should try out VSGAN and with the models from https://upscale.wiki/wiki/Model_Database
I wrote a small guide on how to setup a portable VSGAN when I wrote why it's not included in Hybrid.

Cu Selur

[taigi]

The h264 compression done inside Topaz is out of control

What about exporting to TIFFs and then connecting them with VDub2?

[taigi]

[Attachment 59128 - Click to enlarge]
those parts become very plastic after GaiaHQ. I'm trying to run two not-so-plastic, "weaker" VEAI filters in succession to reach some naturality. And hey, UFO, that's the thing.

[taigi]

BTW, when you enlarge to 8K with VEAI, you have more details, as per my experiment. You may shrink it later.

[Cornucopia]

More details, that's a big NO.


Scott

[taigi]

Why so?

[lollo]

those parts become very plastic after GaiaHQ

If you suffer the same problem, you may try adding some grain (there is a parameter in the options)

those parts become very plastic after GaiaHQ. I'm trying to run two not-so-plastic, "weaker" VEAI filters in succession to reach some naturality.

Please, share the results once done.

What about exporting to TIFFs and then connecting them with VDub2?

Yes, in this way you have control over the compression parameters, but is an additional work and, as you have understood, I am not a fan of Topaz VE AI anyhow, and of upscaling in general. I just wanted to compare Topaz to my "standard" flow to do not give an evaluation without (limited) evidences.

Good luck!

[taigi]

you may try adding some grain (there is a parameter in the options)

Not a trustee of Topaz grain. Maybe later. Need some prove it is good.

share the results once done

There's a possibility I'll share. Would like to have an opportunity to do this easy way. I use this method: yin-yang between sharpness/details and naturality/non-plasticisity.

[taigi]

share the results once done

Sharing VEAI presets that I used last time to achieve not-so-plastic but a bit better details: GaiaCG 200percent and again GaiaCG 200percent. No grain.

[sophisticles]

Some details:

workflow 1
restoration and upscale with AviSynth
- levels and colors correction before filtering
- denoising
- change color matrix
- upscale with nnedi3_rpow2
- sharpening
- levels correction after filtering
- compression with h264 (just to compare and upload the sample)

workflow 2
restoration with AviSynth (as per workflow 1)
- levels and colors correction before filtering
- denoising
- sharpening
- levels correction after filtering
- change color matrix
upscale with Topaz Video Enhance AI
-- avisynth output of previous steps is the input for Topaz

This is not a valid comparison, workflow 1 and workflow 2 do not follow the same order, in W1 you upscale after you change the color matrix but before sharpening yet in W2 the upscale is the last thing you do. You can't make a valid Topaz vs nnedi3_rpow2 comparison under these conditions.

More importantly, and I'm sure that many people will disagree with me, I would think the best results would be to upscale first then do all the restoration that you do.

[lollo]

This is not a valid comparison, workflow 1 and workflow 2 do not follow the same order, in W1 you upscale after you change the color matrix but before sharpening yet in W2 the upscale is the last thing you do. You can't make a valid Topaz vs nnedi3_rpow2 comparison under these conditions.

In W1 I change color matrix and then upscale, as, if I am not wrong, it is supposed to be. I experimented that a sharpening after upscale gives better results than doing it before, so this is my reference flow for this video.

In W2 I change the color matrix and then upscale, as, if I am not wrong, it is supposed to be.
I have no control on Topaz sharpening: as I said I can only guess its "procedures" based on the results, and try to adapt the avisynth sharpening step to do not interfere too much.
I also tried to completely remove the AviSynth LSFmod sharpening step rather than playing with its parameters, but still Topaz is not a clear winner.

If you can suggest a better comparison flow, please do so.

many people will disagree with me, I would think the best results would be to upscale first then do all the restoration that you do.

I am one of them

[krykmoon]

I would think the best results would be to upscale first then do all the restoration that you do.

I say in advance that I'm not an expert at all, this comes from the very limited, short and irrelevant experience I had (doing some experiments)

It really and always depends on source footage, for me.

If I have a very noisy source footage, for example.. or a footage with nasty artifacts.. it's always better to try to remove at least a part of them, before upscaling, or the noise/artifact will inevitably propagate when upscaling... and you will loose some control over them, since you always have a finer control on small details and artifacts BEFORE upscaling than AFTER.

An initial limited/light/controlled denoising (or, for example, deblocking if that's the case) helps a lot if you don't want to propagate nasty artifacts and losing the finer control over them you may have BEFORE upscaling.

[lollo]

Sharing VEAI presets that I used last time to achieve not-so-plastic but a bit better details: GaiaCG 200percent and again GaiaCG 200percent. No grain.

Thanks!

[taigi]

GaiaCG 200percent and again GaiaCG 200percent. No grain.

Thanks!

But again - this is result of a limited experiment only!

[taigi]

@sophisticles: what would you think about upscaling with Topaz Video Enhance AI after all workflow 1? That would enhance some details but would not make everything look plastic probably?

[sophisticles]

@sophisticles: what would you think about upscaling with Topaz Video Enhance AI after all workflow 1? That would enhance some details but would not make everything look plastic probably?

Personally I am a not big fan of upscaling under most circumstances. The only time it makes sense is when you are working with the original lossless or near-lossless footage and after processing you need to upscale the master in order to have it in a standard distributable resolution.

I can't find the page, but Black Magic has a page where they talk about using their Ursa 4.6k G2, shooting in 4.6k, cropping to 3.7k during editing and then slightly upscaling to 4k for the deliverable.

This makes sense and is done all the time as there always something or someone to crop out of a scene.

But what is discussed on this and other forums to me makes absolutely no sense, you're taking content that was already edited, graded, denoised, cropped and mastered, converted into a an end-user friendly deliverable format, often times reencoded again by end-users so they can share it via p2p or streaming site and then you guys try and make a silk purse out of a sows ear by trying to get back to the original master.

And somehow this is seen as a worthwhile endeavor.

As for the original question, you can't enhance what's not there, you can only enhance what is there. That is why i say if you are going to do it, first upscale, preferably to a size bigger than your target yes it will enhance artifacts and blemishes but it will also enhance any details that do exists, then do whatever "restoration" you will attempt and finally downsize to a more realistic target.

So for instance, if you have a 720x480 4:3 dvd, first uprez to 2160x1080, do all the filtering you want, then downrez to 1920x960.

As for where I got these numbers, 720x4/480x3 = 2, 2160/1080 = 2, 1920x960 = 2.

Caveat, i haven't actually done this myself, yet, but that's the first thing I would try.

[lordsmurf]

To be blunt, trying to "restore" Youtube source is generally moronic.

Want better versions? Contact the uploader. Odds are good they'll share the digital pre-YT masters.

[lollo]

But what is discussed on this and other forums to me makes absolutely no sense, you're taking content that was already edited, graded, denoised, cropped and mastered, converted into a an end-user friendly deliverable format, often times reencoded again by end-users so they can share it via p2p or streaming site and then you guys try and make a silk purse out of a sows ear by trying to get back to the original master.

My sample is not. It is a lossles s-vhs capture, restored with a light denoise and a light sharpening (and color/levels correction, not influence). I see no reason to upscale before the light denoise, all the important details are there and are not removed by the filtering. If I do I will only increase the noise together with the details, and this is not suitable, as I have experimented and as others have said.

I just compared the upscale flow in AviSynth and Topaz and the first is absolutely equivalent in my case, so, even if am NOT a fan of upscale, when I need to do it I will use free AviSynth filters.

[sophisticles]

To be blunt, trying to "restore" Youtube source is generally moronic.

Want better versions? Contact the uploader. Odds are good they'll share the digital pre-YT masters.

This is one reason I like Vimeo, sometimes you can download the original source right from them.