Deinterlacing SD footage

Topaz Video AI General
1

I want to upscale various SD videos which were shot on a Sony tape camcorder in the early 2000s. Some are in 4:3 and some are in 16:9. The 4:3 is 760x576@25fps.

My understanding is that, before doing anything in Topaz, I should deinterlace the video without doing anything else to it. Is that correct and can I do this outside of Topaz in a fairly simple and free way? I can then spend time doing that while I wait for any discount on the upgrade of my Topaz version.

Furthermore, I have read about changing the pixels to a square format. I don’t understand that so could someone kindly explain that and, also, how that could be done.

Thanks in advance for replies.

2

I would first try to just enhance the video with Iris Interlaced - I had good results with that approach with quite some older interlaced videos.
Iris does a decent job at deinterlacing content and a sometimes terrific job at cleaning up the video from banding, chroma noise and the like.

Maybe you’re content with what Iris does and so don’t need additional Apps.

Personally I like the V1 results more than V2 because it’s more natural - even though V2 has more details, they are quite often “fake”. Be sure to set the Antialias/Deblur value to a high positive value (>50) and also use a high vlaue for "recover original details (80-85 here).

Se e.g. here for an example of Iris V1 cleaning up an old VHS source and gaining some details while still being very true to the original:

3

Iris added too much wobbling for me to recommend it. (Like edges of buildings and such would wobble when they were supposed to be standing still.)
Handbrake is free and easy to use.
Hybrid is free and more complex to use.

Usually on commercially bought interlaced sources, Hybrid is better.
It was over sharpening some of my home videos, so Handbrake might be the way to go.
Both are better than anything I’ve tried in TVAI.

1 Like
4

My videos I work with are all SD 720x576 PAL from a 1998 Sony TRV-45E Video8 Camcorder (PAR/SAR 16:15, DAR 4:3).

For me personally I found that deinterlacing them in Hybrid and converting to Square pixel outside of TVAI (TVAI also has the option to convert to Square pixel built-in) works the best, before i head off to TVAI for upscaling/enhancing.

To Deinterlace using Hybrid read this.
To covert to square pixel with Hybrid, read this (only the portion that talks about square pixel, the rest is not relevant for you)

1 Like
5

I’ve been using hybrid to prep (if deinterlaced), tvai for scaling/enhancement to prores, and shutter encoder for output to h265 using crf.

pita but great results.

6

Thank you for all of the helpful replies and I will try those out.

7

Many thanks. I have downloaded Hybrid and used your guide to deinterlace the video.

Having done that, what would you then recommend as the model (or workflow/steps) to use in Topaz to upscale? I am assuming that it is only worth upscaling to HD as upscaling to 4K would be pushing things too far.

8

I don’t know what source you are using, but my flow for my source videos is run a Nyx v1 (Auto) clean up (keeping it at the same resolution), because mines a bit noisy. then upscale x2 using Dione TV.
For me it comes out sharp relative clean of artifacts, much details preserved & it keeps the original looks of the video, just in HD resolution.

I know @ssbroly did extensive testing to the border line of obsession (in a positive way) for it to become perfect , so maybe he has better method that might suite you better.

Also @ForSerious developed good results and has some workflow(s), maybe he can share as well.

I am sure there are others here as well, but those are the two that i was more exposed to on upscaling workflows.

P.S. yeah, going from SD → 4K might be a disappointment, I would stick with SD → HD until you get good experience with the tool at least…

9

It’s true that I tried a lot of methods! lots of models, different workflows, changing the order of filters with hybrid, all this while testing to produce in sd 720x576, in sd 768 x 576, in hd, before sending to hybrid, iris, iris + artemis hight, artemis hight + ris, artemis medium + iris, artemis hight + iris, artemis hight + gaia + iris, all that in HD, SD etc… I will try to be brief about my results, which helps me seems to be best for my SD videos, it is to use qtgmc with hybrid everything at “placebo” and above all leaving “final temporal smoothing” at 0. if we put 1, 2 or 3, the image is better , less noise, artifacts etc., but in areas with movement, details become unrealistic, even with 1 as a value. So for me it’s because I’m very picky and I really wouldn’t want to lose any details, but if someone agrees to lose a little detail in the movements, the value of 1 is good. 2 or 3 is too much, and it will clearly lack detail… then in hybrid, I use a dehalo filter, in particular to remove a bit of the white halo from SD videos, something that Vai does not do, a filter by in relation to the color to bring out the dark areas, it is details to recover later with vai. I did some tests with the “nnedi3aa” line filter to smooth out the edges a little, it’s not bad, but when you apply vai models, some scenes can make details a little unrealistic like vegetation, or stones on the ground. especially when you upscale to HD at the end. the other line filters like “eedia” or “maa” smooth the edges even more, so even a little more unreal. the “aaf” filter may be very good, I don’t know exactly how to use it but we can configure it quite extensively. so I think I will continue to experiment with it. I tested quite a bit without a line filter because I tell myself that with the arrival of iris v2, the edges are pretty well repaired overall and if we send a good video processed with hybrid, there will be no loss of detail. the problem is that iris v2 is very good at restoring as much detail as possible, when we produce from SD to SD (without upscaling) there are sometimes strange artifacts which were not there before… I have noticed that when switching from sd to hd, these artifacts are less visible, this is why for me the best setting is to use artemis hight (recovery original details 100) in sd, then to remove the last artifacts, perhaps it will be necessary to use artemis hight + proteus, or the opposite, and then I strongly advise using iris v2 for the upscale in hd, but you have to make sure that it I don’t have too many artifacts anymore, I also tried nyx + artemis hight, nyx + iris, artemis hight sd + iris hd + gaia hd. the gaia model adds a lot of detail or it accentuates the existing details perhaps but it does it very well because the video appears more realistic but what blocks it at this level is that the few artifacts which slip through the mesh of the net per iris, they are filled, and therefore sometimes we have strange artifacts. Anyway, I think a good base to try is artemis medium 100% in sd + iris v2 in hd. I can’t detail all my tests, for the moment, I’ve stopped a bit and I’m waiting in the hope that iris v2 will be just a little improved so that it no longer produces these few strange artifacts at times, and then for me v4 has a good interface but too buggy like most people think, just for the preview, I can’t work with it… I only have a gtx 1660 so not the best gpu faster too haha whatever, after all the passages with vai, there is always a little noise, something that I adjust with vegas pro with another filter to finalize (colors, dehalo, sharpness, grain at the end, and above all: neatvideo as the first filter!) I think that the secret to properly upscale a video is to improve it in its original resolution, the fewer artifacts there will be, and the cleaner the video will be when we will put it in HD. the most complicated thing is to improve it, removing noise and artifacts without removing details. on the other hand @akila, I will try to refine your method with dione dv, I had already tried, but I found that dione dv, dione tv etc remove a lot of details… so maybe depending on how I apply my filters with hybrid before, the details I can keep, I don’t really know!

10

If you do, let me know on your findings and I’ll see if it works for me and I’ll adopt it as well.

11

Okay, I’ll try to find time, at the moment I have less because I’m renovating my apartment so I don’t use vai too much anymore, but if ever I see that a future version can be improved compared to the current version, I’ll try anyway :wink:

1 Like
12

Thank you for such a detailed reply from you and @Akila . I am very much a novice so am in no position to express a view on what is best. But I will try the workflows that you have outlined. There are a few things that strike me (just observations really).

  • One is the use of TVA on footage to improve/enhance it without changing the resolution. Indeed, as you say, you should do as much improvement without changing the resolution as you can and then upscale “as late as possible”.

  • Second is that there isn’t a definitive workflow for SD->HD. Of course, this is because different people will have different source footage. But, for someone still at the very start of the skill ladder, the various TVA options can be confusing. It isn’t clear (it may be my lack of understanding) exactly what each model in TVA is designed to do. It would be good if that were set out somewhere (or linked if it exists and I didn’t find it).

  • Third is that the forums are not that easy to navigate to find information like that above. Which is why I am so grateful to you @ssbroly and @Akila especially for your detailed replies.

So, to summarise, as I understand it, the workflow to try out goes something like this:

  1. Deinterlace using Hybrid (SD->SD)
  2. Convert to square pixels using Hybrid (SD->SD)
  3. Use a dehalo filter in Hybrid (try nnedi3aa) (SD->SD)
  4. Use Artemis Hight in TVA with recovery details 100 or Artemis Hight + Proteus (SD->SD)
  5. Finally use Iris v2 to upscale (SD->HD)

Once again, thanks for the help.

13

in fact yes, it is better to improve without upscaling because when you upscale, the artifacts are also amplified and it is harder to get rid of them. for vai models, the best is to try on your own sources, after deinterlacing with qtgmc, artemis medium can be enough to apply after iris v2, the only problem is that artemis medium removes some details in dark areas and loses a bit more detail than artemis hight, but it smooths edges better and therefore removes artifacts better compared to arrtemis hight. It depends how picky you are about the ratio of details retrieved. It’s true that the navigation on the forums isn’t great but we always end up finding what we’re looking for :wink:

for hybrid, converting to square pixel (768x576) or leaving it in original pixel (720x576), it doesn’t change much in my opinion, it gives me the same image quality. when you change the resolution of a video, you also have to choose the upscaling method (lanczos, bicubic, bilinear etc) lanczos is known to be of better quality, but with an old workflow that I had (still with hybrid and vai), bicubic or bilinear gave me better results.
we also don’t have to use a halo filter with hybrid, because vai has one with proteus and iris, but on the other hand if we want to remove the light halo around the edges, yes you have to do it with hybrid. nnedi3aa helps reduce anti-aliasing and reduce some artifacts, but it is not without quality consequences. the image will look slightly less realistic. but you really have to be picky about the quality because even for me who is, I can’t just see the difference when playing the video.