Search options (Page 2 of 2)

Yep, and also there appears to be a drop shadow underneath the top row of brick of the water fountain - I think that might be a field alignment artefact?  QTGMC is smoothing it out.

My QTGMC clip was just QTGMC only (nothing else, no other processing like TIVTC) with preset="slower", sharpness="1.0".

My detelecine clip was TIVTC only (TIVTC settings in your GUI screenshot).  I'll try to translate any of my settings which are different to your screenshot:

TFM mode: 2 way match
TFM advanced tab: combing threshold 9 (incorrectly labelled as chroma threshold in GUI)
TFM advanced tab: MI 128
TFM advanced tab: scene threshold 100
TDecimate: hybrid = enabled  *
TDecimate: VidDetect A or B  *
TDecimate: VidThresh 4.5 *
TDecimate: obey TFM hints = false *
TDecimate: Chroma unticked
TDecimate: Denoise ticked (this doesn't denoise the output image, it just does it internally to get better frame metrics)

* if all 4 of these can't be set, then the "hybrid" frameblending feature probably won't work properly, and in this case set hybrid = disabled

In your screenshot it's got QTGMC as the deinterlacer for TFM.  That just means if after the detelecining process there are any combed frames still coming out of it, those will get deinterlaced with QTGMC.  This is needed as there will be the odd sequence here and there which can't be detelecined due to the weird editing they did back in the 90's where they'd often add purely interlaced frames (which can't be detelecined) for the special effects or when splicing in certain scene changes.  The problem is that you get false positives like the Jeep grille etc. which look like interlaced frames to TFM, so those frames end up getting deinterlaced when they could have been detelecined instead.  That's why I set the combing detection threshold higher (MI=128) to try and avoid those false positives, but they still can happen.

Also keep in mind that app you're using appears to be using the Vapoursynth version of those filters, whereas I'm using the Avisynth version.  They should be pretty much identical though, but you never know.

No upscaling, except from 720x480 -> 720x540 to convert it to square pixels 4:3 (720/540 = 4/3).

The full script I used for that file was:

# set up source clip
clip = "C:\S01E04.mkv"
video = LWLibavVideoSource(clip, stream_index=-1, repeat=true, cache=true)
audio = LWLibavAudioSource(clip, stream_index=1, cache=true)
AudioDub(video, audio)

# QTGMC deinterlacing
cores=4 
QTGMC(preset="slower", FPSDivisor=1, EdiThreads=cores/2, Sharpness=1.0)

# scale to 4:3 square pixels
LanczosResize(720,540)

# turn on multithreading for performance increase
Prefetch(cores)

Yeah it would be interesting to see what Topaz can do with that file I uploaded.  ireactions mentioned deinterlacing artefacts, but the file I uploaded is progressive, so as long as Topaz is configured to not do any further deinterlacing or detelecining, it shouldn't introduce any new artefacts in that regard.

There is an explanation of how it works here: http://avisynth.nl/index.php/QTGMC#Algorithm_Details

On a synthetic test pattern it actually resolves less resolution than detelecine because it doesn't weave fields to progressive frames; it only interpolates them in a very clever way: https://imgsli.com/MTY0MzIx

Note the alternating 1px white and black lines become solid white/black (since an interpolation of white px above and white px below = white) and the converging lines in the circle start to become slightly blurry the closer they get together. 

But that is by design, and season 1 doesn't have a full 480p to resolve anyway due to the field alignment issue, so it is my intention to blur those lines together.  But this is why I wouldn't use it on the later seasons which don't have the field alignment issue - those will have better resolution using delecine instead of QTGMC. 

But, subjectively I still think QTGMC looks very nice in its own unique way.  I especially like that it outputs 60fps as this preserves all the original frame pacing.  You may have noticed the animations in the intro sequence are smoother, like the spinning earths, and a couple of brief 60fps animations at other points.  With detelecine those would get decimated down to 24fps resulting in stutter or frame blending artefacts of those particular animations.  The rest of the show itself is almost always 24fps though so no improvements there - actually QTGMC has slightly worse frame pacing on 24fps sections as the output is 60p which means 3:2 judder.  But I don't really mind that, actually I find it interesting and imbues a certain quality.   edit: and I just remembered QTGMC's output can be converted to 24fps if desired, using its FPSDivisor=2 param which will output 30fps, and then TDecimate can be used on that to remove duplicates to 24. 

The 4 slowest modes are: slow, slower, very slow, placebo

The default is "slower", and is what I used for that file.  My i7-4790k (4c/8t @ 4.0ghz) can just barely manage "slower" in realtime (~65fps... so 5fps headroom) but "slow" seems to look pretty much as good and gives some nice headroom at about 90fps so I use that when screening an episode.

I'd imagine ireactions can give the best answer here, but I'd guess Quinn is supposed to be in his mid 20's and the professor in his 50s?   Under this assmption, 2023 Quinn would be approximately the same age as 1995 Arturo.

https://i.lensdump.com/i/THFfAP.png
https://i2.lensdump.com/i/THFFVm.png

In motion the pattern changes every frame like film grain - is that what you're seeing too?

Weird how they made animation through film back then, I can't even imagine how that would work.

Geoblocked by CBS

I think I figured out what MCH is using to smooth out those tree branches at top of frame - this popular deinterlacer called QTGMC. As I am able to replicate the same image quality with QTGMC.

What I think is happening: MCH has set the post-telecine combing detection threshold lower than I have, and as a result of this many frames in that scene are being falsely detected as combed (interlaced), causing QTGMC frames to be substituted for those frames, thereby inadvertently smoothing out the artefacts which are baked into the source.   Whereas my script has combing thresholds set much higher so mine is just weaving fields (detelecine) to progressive frames and showing those frames as is, warts and all. 

The reason I use a higher combing threshold is due to false positives, things like car radiator grilles or clothing with pinstripes can cause the frame to be flagged as combed (interlaced) and trigger QTGMC (or whatever deinterlacer I choose) which doesn't extract as much resolution as detelecine (weaving fields to progressive frames).

By the way QTGMC is effective at suppressing the field alignment issue too, and looks sharper [1][2][3] but introduces moire artefacts on stippled patterns [4].

Yes, it's because the person who did the encode (MCH) cropped the small ~8px black pillarbox bars down the right and left side, then set the video stream's aspect ratio flag to 4:3 which causes the directshow video renderer to stretch it back to 4:3 at runtime, creating a 1.022% (720/704) wider aspect.   I can correct that manually but it's trivial.

I believe the source frame for both images is the same, because I stepped through 1 frame at time and counted them manually - 67 unique frames for that opening street scene in both mine and MCH's, and we are paused on frame 4 of 67.  I also inspected neighbouring frames and those are all clean in MCH's vesion too.   So whatever MCH has done is some kind of magic that I don't understand.

I set my detelecine filter to debug mode with TFM(display=true) and none of the frames were detected as interlaced, so they aren't being deinterlaced by TFM, therefore TFM is not causing those artefacts.  And I disabled TFM's deinterlacing anyway with TFM(PP=0) so I'm definitely not adding those artefacts. If I do absolutely no video processing at all, the same artefacts are still present, so I am confident they are baked into the source.   

I have no idea how MCH removed them.   Maybe they were removed by fluke as part of the "remove 1 in 5 frames" when decimating the field matched 30p to 24p.  Perhaps that 1 in 5 just happened to land on those problematic frames and remove them?   I tried TDecimate(mode=0/1) which toggles between removing "most similar duplicate" and "last frame in string of duplicates" without improvement.  For this street scene the camera isn't moving very much and all of them are below the CFrameDiff threshold so almost all of them are detected a duplicates, and it has to remove 1 in 5 from the field matched 30p stream created by TFM.

Well, mostly out of curiosity to see if it could be fixed, and then later because we want to also fix those problematic Universal NTSC episodes as well. It seems there are episodes where the field alignment issue is not visible in the end credits text, but IS visible in the live action shots.   Perhaps because the credit fonts are serif'd in such a way that hides it a bit.   The end credits for season 4 use a much thinner font so they won't be able to hide there.  The Mill Creek version has much worse alignment and it shows up on all seasons end credits.

Thanks, yep I would expect the Universal releases to all be the same MPEG2 video streams, just as they are for the 2 PAL Universal version's I've got.     

I'm not sure what's going on, either there are multiple Universal NTSCs with different image quality, or whoever did this particular transcode of it is wielding some voodoo magic video processing that I cannot get my head around.

Yep it's pretty bad.  Here's my attempt at cleaning it up so far...

Scene 1 - city square
Scene 2 - forest
Scene 3 - studio
Scene 4 - end credits

To see the actual quality you'll have to download the files and play them with a media player.  The version that plays embedded in your web browser is Google/Youtube's transcode of it, not the original file.

Filters used in this order:

DeCross (reduce analogue rainbow dotcrawl artefacts)
Detelecine to 480p
Deflicker filter (reduce field alignment drop shadow lines)
Convert colourimetry to HD Rec.709
4x AI upscale to 2880x1920p
Downscale & sharpen to 1440x1080p
Slight colour saturation increase

I played around with noise reduction but didn't like the result.  Might have to play around with the thresholds a bit more.

I've had some time to view the Season 1 Mill Creek episodes on the big screen with the deflicker filter applied, and I think with this filter it is in watchable condition.  It is not good quality by any means, but I think it just passed the threshold from unwatchable to watchable.  The drop shadow lines are absolutely gone.  The weird facial features on actors at a distance are completely gone.  The image certainly isn't as sharp, but it's not out of focus or difficult to look at either.  I am adding a conservative amount of Lumasharpen through the media player's renderer to compensate for the deflicker filter. 

This is the Avisynth script I'm using:

# setup source video
sourcefile = "D:\Video\Series\Sci-Fi\Sliders (Mill Creek DVD)\Season 1\S01E04.mkv"
video = LWLibavVideoSource(source=sourcefile, stream_index=-1, repeat=true, cache=true)
audio = LWLibavAudioSource(source=sourcefile, stream_index=1, cache=true)
AudioDub(video, audio)

# detelecine: field matching to 30p
TFM(mode=0, slow=2, PP=6, cthresh=9, MI=96, scthresh=100, 
\    clip2=bwdif(last, field=-1, thr=2, pass=true))

# detelecine: remove duplicates to 24p
TDecimate(mode=1, cycle=5, cycleR=1, hybrid=1, viddetect=2, vidthresh=4.5,
\    chroma=false, denoise=true, hint=false)

# deflicker filter
nnedi3(field=-2)
Merge(SelectEven(), SelectOdd())

# aspect ratio 4:3 square pixels
LanczosResize(720, 540)

# optional: raise audio volume to highest peak without clipping
# Normalize(1.0)

How to use:

1. Use MakeMKV to dump your original disc to .mkv files.  The Avisynth script will not necessarily work properly on other people's rips, it needs the original interlaced video stream from the disc.

2. Install the latest version of Avisynth+ 64-bit

3. Download the latest version of the filters used by my script: LSMASHSource, bwdif, nnedi3. Drop the 64-bit .dll files into eg. C:\Program Files (x86)\AviSynth+\plugins64.

4. Paste the script into a text file, save it as an .avs file and open it with a media player that supports playing Avisynth files. I'm using MPC-HC with MadVR as the renderer, which can be downloaded as a preconfigured pack here.  To avoid audio sync issues with Avisynth files I had to set MPC-HC to use MPC Audio Renderer instead of the default Sane Audio Renderer (Options > Playback > Output > Audio Renderer > MPC Audio Renderer) .

Great post - I enjoyed reading it.

In my view the limitation can function as its strength.  An example might be retro video games - why would anyone play around with simple blocky 2D graphics when you can have massive 3D worlds?   Because those limited graphics and sound have a certain charm to them.   

The way I see it, the 16:9 frame is like a canvas and you can put whatever you want in there.  Framing and composition within the 16:9 frame doesn't necessarily have to be 16:9.   Again, I'm not proposing to put black bars down the sides, I'm only talking about the composition/framing of shots.   What if the director wanted to use a wider aspect like 1.85:1 - is this allowed?

But ultimately it comes down to this: I've never seen a remake that wasn't a shadow of its former self.

Yeah I agree the general public won't accept 4:3, but I don't think they will notice if there are no black borders.   All it does is change the artistic framing/composition of scenes, eg. actors will be standing in the middle of the frame instead of being spread out over the full width of the frame.   

I'm watching Charmed HD remaster (1995 season 1) and it looks to be 4:3 safe with framing of the shots and text overlays being kept within the centre 4:3 zone.  I think it was mostly in that transition period in the 2000's when half of viewers didn't have 16:9 TV's yet so they shot stuff as 4:3 safe to ensure everyone's screens were filled.   

Does Sliders looks cheap and dated to you?   Certainly the video quality of the DVD's is cheap and dated (especially season 1 and 2) but if it was rescanned in HD that would still retain the "90's TV show production values" imo.

Yeah I noticed that too.
Here it is vs Universal NTSC : https://imgsli.com/MTU4MDU2

My screenshots are showing my intentional deflicker smudging filter applied after I've already detelecined.  Detelecine itself shouldn't produce any smudging as it mostly just performs field matching to weave as many field pairs as possible so that you get true 480p resolution frames coming out.   Although I've seen a problem where the frames that come out of the field matching process are double checked for combing in case they contain fields that couldn't be matched to other fields ("orphan fields") and if Handbrake doesn't set the thresholds right (or disable the decombing feature entirely) it can cause false positives.  eg. in this clip (played by your media player not in web browser) at around the 9 second mark notice the aliasing & line twitter on the blinds and upper walls - that would be the decomber falsely detecting those pixels as combed due to the stippled patterns in the blinds.  Other patterns like actors wearing striped/checked shirts, car radiator grills or the rims of a stack of plates in a kitchen scene can set off the decomber too so I like to disable decombing entirely for clean sources, or set the threshold very high so it only triggers when there's lots of combing, or only check n frames either side of scene changes (as this is mostly where orphan fields end up due to the editing equipment in the 90's not making sure every scene ends perfectly at the end of a 3:2 sequence).  Another issue is repeated frames like the one at 11 seconds, that can also be mitigated with the right decimation settings as well.   This is why I want the original discs so I can do my own real-time Avisynth detelecine with my own optimal settings rather than trusting Handbrake or someone else to get it right.

I've been spending some time with the PAL Universal version this evening and I think it's reasonable quality - it's quite a lot better than Mill Creek, on average (season 1 is still a bit meh).  Since Universal NTSC only has a few problematic episodes, perhaps it might be better to just replace those few problematic episodes with the PAL version (detelecined to 25p and then slowed down to 23.976 of course to match the NTSC version, which is trivial to do in Avisynth).   

Regardless, I still want a solution to this issue and it looks like there is a neural net based smoother that works wonders on animation with this field alignment problem, but it might not work on live action scenes very well and I haven't figured out how to actually use it yet (it seems to require python so it would involve reencoding every episode which I really want to avoid).  I'll report back with any findings.

Is that Universal or Mill Creek?  Sounds like the saturation may be higher for those later episodes (hard to tell with Mill Creek since it all looks like crap).

Regarding the field misalignment issue, it is possible to smooth it out a little by applying a kind of deflicker filter which works by blending each pixel with the pixel above/below.  Here is what it looks like, not sure if you guys would like it

https://imgsli.com/MTU3OTQ2
https://imgsli.com/MTU3OTQz
https://imgsli.com/MTU3OTQ0  (looks like Quinn has a mustache & Wade has 3 lips)
https://imgsli.com/MTU3OTQ1

I'm not sure if it's the lesser of 2 evils as it does smudge a lot of resolution out of the image.

Does anyone know where to buy Season 5 on Universal NTSC DVD?  I can't find it anywhere, even a used copy.   I've already bought the Mill Creek NTSC release but it has the field alignment issue.   I've bought the PAL DVDs but could only get seasons 1-4.   PMs very welcome!

btw I did some experiments in Photoshop and can reproduce the Muck Creek field alignment issue exactly.  Trying to figure out a way to reverse it in Photoshop, then once I know it works in theory it should be translatable to an Avisynth script.  But I'm still not sure if the damage is done in a way that is reversible or not.   It's kind of like lossy vs lossless compression - one of them can be reversed to get a 1:1 copy of the original, the other cannot.   Reversing the field alignment issue may still be lossy.  I think if I knew the exact scaling factor and algorithm they used for the digital zoom which put the fields out of alignment, then maybe I could reverse it 1:1.

I'll add my comment: I do not like it when movies are intentionally color graded in a different way for a new release.  To me it's like taking someone's music and putting a U-shape EQ and compressor on it, and thinking you made it better and that everything is fine.   Then to add insult to injury, they tell you the U-shape is actually "more accurate" because it compensates for people's hearing, or lower listening volume levels, or some irrelevant nonsense that has nothing to do with the original artistic intent.   Changing the source material is wrong for the same reason it's wrong to crop 1.85 movies to 16:9, or zoom in on 4:3 a little bit to reduce the pillarbox bars.    In some cases you can even preserve the artistic intent, but then it turned out that wasn't what got shown on screens anyway cause the equipment they used was adding its own nonlinearities which produced a different look and feel to what the director intended, which then got reified with the movie.   Warner Bros seems to be a particularly bad offender in this regard with many of their backcatalogue being redone in orangeteal with crimson reds, eg. Lethal Weapon.

Well, I remember reading a while back that the MacGyver negatives were lost and that's why there would never be a HD remaster.  Then in 2018 CBS announced they had the film negatives and were remastering it.  So you never know, there could be negatives of Sliders in storage somewhere.   And if there wasn't, there may be copies floating around elsewhere kind of like what happened with Project 4k77.   

The big issue is the time and money taken to scan them in and re-edit all the episodes to splice in the new film frames alongside the frames with CGI which were probably rendered directly to video at 480p.   Kind of like what you mentioned with X-Files, although I remember reading the X-Files HD was done on the cheap with some clever software to automate the colour grading process and it ended up costing less or something.

Another possibility is that there could be "broadcast quality" SD masters sitting in the archives of TV networks.

Just to clarify: at first that's what I thought was happening, but now I think it's a field alignment issue caused by the fields being scaled independently of eachother before being weaved back together.  This effect is "baked into" the source and I'm not sure there's any way to reverse it.

I believe the technical term is "generation loss" - every time it gets re-recorded a bit of quality is lost.  Of course, you could compensate by setting the bitrate very high.   

In my setup I'm using MakeMKV to "remux" the MPEG2 stream on the disc into a mkv file, which doesn't cause any generation loss (it's just copy pasting the MPEG2 stream from disc to mkv file).   Then to play it I write my Avisynth detelecine processing options in a text file and save it with a .avs extension, and open the .avs file with MPC-HC media player and the Avisynth processing gets done in realtime.   This is the highest quality method as it avoids transcoding to another format like x264 which results in some generation loss  (unless you set the bitrate very high... for me it's mostly a convenience thing as encoding video takes a lot of time with my CPU pegged at 100%, and if I want to change something in my Avisynth script, I'd have to re-encode the whole lot).

Well the 2013 Via Vision PAL DVD is highly disappointing.  In terms of clarity and colour it's about half way between Mill Creek NTSC and Universal NTSC.   Here are some screenshots which are zoomed in 200% nearest neighbour (duplicating each pixel into a square of 4 pixels)...

Scene 1: Mill Creek NTSC , Via Vision PAL , Universal NTSC
Scene 2: Mill Creek NTSC , Via Vision PAL , Universal NTSC

Clarity and colour seems closer to the Universal NTSC release than Mill Creek, but there is another huge issue with the PAL version - the frame rate is inconsistent.  Most of the scenes are sped up to 25p - this would have been fine as it's easy to slow back down to 23.976 in realtime with Avisynth using AssumeFPS(24000, 1001, true) - however many scenes for some inexplicable reason were done at 24p frameblended to 50i, so there is no possibility of detelecining these scenes, and there are a fair few of them.   It sounds like they didn't pitch correct the 25p sections either so the actor's voices get noticeably deeper on those scenes.   

Still have the 2008 Universal PAL release coming in the mail tomorrow, but not getting my hopes up.  I'd anticipate it's exactly the same MPEG2 stream on the 2013 Via Vision PAL release, just with older packaging.   I'll probably chuck the whole lot back on ebay and burn the Mill Creek discs.

edit: managed to scoop a used copy of Universal NTSC off eBay with international shipping - expected delivery date end of March.  Also spotted this rare item, but not sure which release it is exactly.  Probably Universal NTSC?

Cheers, I just ordered this which appears to be Universal's PAL release.  There were no NTSC versions listed in my region of ebay which I would have preferred.

I suspect the PAL version might actually be NTSC video on the disc though, as I've recently bought DVDs of other US shows that say PAL region 4 on the box but the MPEG2 video on the disc is still 720x480i, eg. The Ray Bradbury Theater & Caroline in the City.

edit: hmm, it seems there are actually 2 different PAL DVD releases in my region: one published by Universal in around 2008, and another published by Via Vision in around 2013.   I am not sure if they both contain the same MPEG2 video streams, so I bought both just to make sure.   I suspect the Via Vision one may be the same video contained in their box set, although the box set has a release date of 2019 which is much later than 2013 so maybe not.   My PAL/R4 versions of Ray Bradbury Theater and Caroline in the City are both Via Vision releases and contain NTSC MPEG2 video streams, for whatever that's worth.

Actually I just realised I've seen these type of double line artefacts before. It can happen when interlaced video is scaled without deinterlacing.   i.e just taking each field of 720x240 and resizing them individually, then weaving them back together into a progressive frame.   It creates this "mice teeth" effect because the fields were resized independently of eachother and don't fit back together quite properly.  I observed this on a recent Bluray release of the show Fawlty towers - https://forum.doom9.org/showpost.php?p= … stcount=71

So my hypothesis is that Mill Creeks version went through this field scaling process at some stage which offset the field structure slightly.   In the screenshots you can see it is zoomed in slightly compared to Universal, maybe that was when it happened.

Hmm, not sure what's going on. 

Here is the same frame Universal vs Mill Creek:

I am quite confident this is not a deinterlacing or detelecining artefact, as both have been field matched to weave to a progressive frame.

edit: this is from the end credits of S01E03 Summer Of Love