Transformers 80's cartoon video restoration

[Sunstreak22]

Hello, I would like to improve the video quality of the original 1984 Transformers cartoon from NTSC DVD release. I would like it to be as close to the creator's intention as possible, ensuring that the lines are neither thicker nor thinner than originally drawn and that details are preserved. The first problem I encountered is dealing with telecine. From what I have checked, this release has hard telecine. TFM handled most of it well, but in some segments, there are noticeable jagged edges (e.g., S01E01 - from frames 90 to 209, S01E02 - from frames 375 to 456). In the screenshots below, I have highlighted example areas in red circles where this issue is visible.


[Attachment 80596 - Click to enlarge]


[Attachment 80597 - Click to enlarge]

What causes these jagged edges, and how to correct it? I am using the following simple AviSynth script:

Mpeg2Source("C:\Users\PC\Videos\S01E02.d2v")
#ColorMatrix(mode="rec.601->rec.709", interlaced=true)
AssumeTFF()
TFM()
TDecimate()
Crop(8,0,-8,-0)
#nnedi3_rpow2(2, cshift="Spline64Resize", fwidth=960, fheight=720)
#Spline64Resize(2880,2160)

I plan to watch this on a 4K TV from a connected laptop and open the script in MPC-HC to achieve better scaling than the TV. I'm also considering using nnedi3_rpow2 to reduce aliasing, but it sometimes tending to merge small gaps between adjacent animation lines. How can aliasing be removed in this case without losing details?

I would be grateful for any help and explanations.

[timtape]

Both video samples show film artefacts so the cartoons were probably originated on motion picture film as many cartoons were. The DVD copies will have degraded the films' resolution so they would not be the ideal sources if the aim was a full restoration.

[Selur]

Not without losing and gaining some details, here is what BasicVSR++ and NNEDI3 do:
Images: https://imgsli.com/Mjc4MjQx Script: https://pastebin.com/9yy1gPnn
Videos attached. (the result can be improved some more; I just wanted to see what BasicVSR++ would do)

Cu Selur

Ps.: regarding the aliasing, using something like:

clip2clip = havsfunc.QTGMC(Input=clip2clip, Preset="fast", TFF=False, FPSDivisor=2)
clip = core.tivtc.TFM(clip=clip, clip2=clip2clip)
clip = core.tivtc.TDecimate(clip=clip)# new fps: 23.976

will take care of it.

[poisondeathray]

The jaggy aliasing is from TFM deinterlacing when it shouldn't be - it detects some horizontal lines in the picture and it registers as "combed", so it applies deinterace .

Either adjust the settings so it's less sensitive (look at the instructions) , or to be drastic use TFM(pp=0) to disable post processing completely, but if you have orphaned fields somewhere, you can get combed frames. To debug, use TFM(display=true) and adjust the settings

[Sunstreak22]

Both video samples show film artefacts so the cartoons were probably originated on motion picture film as many cartoons were. The DVD copies will have degraded the films' resolution so they would not be the ideal sources if the aim was a full restoration.

Unfortunately, this animated series has only been released on DVD. From what I've found, the master was created from 35 mm film to Digital Betacam videotape. The studio that handled this described the process at the beginning of this video: https://youtu.be/qvu_DL0YHd0?feature=shared&t=11

Not without losing and gaining some details, here is what BasicVSR++ and NNEDI3 do:
Images: https://imgsli.com/Mjc4MjQx Script: https://pastebin.com/9yy1gPnn
Videos attached. (the result can be improved some more; I just wanted to see what BasicVSR++ would do)

Upscaling using BasicVSR++ looks interesting and promising, but currently, I prefer to minimize the use of AI or not use it at all because it looks too fake to me. Film grain doesn't bother me, and I prefer to keep it to avoid losing details. Thank you for the script, I will check what can be achieved with it.

Either adjust the settings so it's less sensitive (look at the instructions) , or to be drastic use TFM(pp=0) to disable post processing completely, but if you have orphaned fields somewhere, you can get combed frames. To debug, use TFM(display=true) and adjust the settings

From what I have noticed, even the default settings do not always detect combed frames in darker scenes. Would the best solution be to split the video into segments where I notice incorrect detection and adjust the settings for those segments?

[poisondeathray]

From what I have noticed, even the default settings do not always detect combed frames in darker scenes. Would the best solution be to split the video into segments where I notice incorrect detection and adjust the settings for those segments?

You can do that and/or specify ovr for the overrides file in TFM - there are examples in the documentation for the syntax

[SaurusX]

I read through the other responses. First let me say that I have extensive experience with this show and no one has mentioned the true causes here.

One, these are old cartoons from film with gateweave and TFM will miscombine fields like in your examples. The jiggliness in the video will confuse it and it will pull the wrong combination thinking it's OK, because there is no combing. if you look at TFM with "display=true" you'll see this is the case in almost all instances. The only way to fix it is with a manual override file.

Two, for this show you need to have "mode" in TFM set to 4 or 5. There are some bizarre field combos that come up and it needs more options available to search.

Good luck!

[Sunstreak22]

You can do that and/or specify ovr for the overrides file in TFM - there are examples in the documentation for the syntax

Thank you for the advice, I will definitely use the option with an override file.

Two, for this show you need to have "mode" in TFM set to 4 or 5. There are some bizarre field combos that come up and it needs more options available to search.

Good luck!

Thank you for the tips, I'll see what I can do. For the first three episodes, I didn't notice any differences between the mode set to 1 or 5, so I preferred not to risk increasing that parameter, but I will do more tests. Can setting the mode to 4 or 5 cause jerkiness in some parts of the episodes or any other issues in this case?

[SaurusX]

Thank you for the tips, I'll see what I can do. For the first three episodes, I didn't notice any differences between the mode set to 1 or 5, so I preferred not to risk increasing that parameter, but I will do more tests. Can setting the mode to 4 or 5 cause jerkiness in some parts of the episodes or any other issues in this case?

Oh it does make a difference. Too low of a setting actually will cause jerkiness most noticeable in scrolling scenes. A small hitch will creep in every now and then. I've only ever had positive results by increasing the mode to 4 or 5. I think with 4 it'll be near perfect, but this source is whack.

[Sunstreak22]

One, these are old cartoons from film with gateweave and TFM will miscombine fields like in your examples. The jiggliness in the video will confuse it and it will pull the wrong combination thinking it's OK, because there is no combing.

Right, from what I have noticed, incorrect frames are also created due to poor field matching. As a result, a frame that should be repeated once is repeated twice or not at all. Setting in the ovr file which frames are not combed in this case is not sufficient, in some parts marking which fields to combine is also necessary.

If I understand correctly, when viewing the original unprocessed file and a frame appears progressive, the current fields will most likely need to be matched tugether (c), and when frame is combed, a different field should be used for matching (p or other option). The typical pattern is cccpp, but when orphaned fields appears or there is little movement, the situation becomes more complicated. Is there a way to verify whether the fields have been combined correctly, how they should be combined, or can this only be assessed visually after the fields have been matched?

[Selur]

it can "only be assessed visually after the fields have been matched"

[lordsmurf]

Some VHS are different source. Just use better replacement of bad sections, edit new video. I hate these issues on the 1st trio of episodes. I can help with at least episode 1. Do you want to try that? I have most VHS releases, this is the reason why. Let Selur upscale if he's willing.

[Selur]

Upscaling using BasicVSR++ looks interesting

Just for clarification: I did not use BasicVSR++ for upscaling. The upscaling was done by NNEDI3. BasicVSR++ was used for cleaning the source.
There are way too many options one could use for filtering and upscaling depending on once own preferences,... (a few examples: https://imgsli.com/Mjg3NTIw)

Cu Selur

[Sunstreak22]

it can "only be assessed visually after the fields have been matched"

Okay

Some VHS are different source. Just use better replacement of bad sections, edit new video. I hate these issues on the 1st trio of episodes.

Examples such as those I sent in the first post can be fixed as poisondeathray and SaurusX mentioned by using an override file. I made an ovr file for a section of the first episode and I managed to fix those jagged edges (I'll send the ovr file and write more about it later). However, there are other sections that will require replacement for various reasons (A film reel with unfinished animation was used, it was poorly edited, etc.). To fix these, different sources like VHS releases are needed either for replacement or as a reference. Probably all VHS releases of this cartoon use the original broadcast masters, DVD releases don't always do so, which causes various issues.

I can help with at least episode 1. Do you want to try that? I have most VHS releases, this is the reason why. Let Selur upscale if he's willing.

Any help is welcome . Yes, I would like to try that. I will definitely replace some sections with another source.

Just for clarification: I did not use BasicVSR++ for upscaling. The upscaling was done by NNEDI3. BasicVSR++ was used for cleaning the source.

Sorry, my mistake. I’ve never used BasicVSR++, I thought it was only for upscaling.

[lordsmurf]

(A film reel with unfinished animation was used, it was poorly edited, etc.). To fix these, different sources like VHS releases are needed either for replacement or as a reference. Probably all VHS releases of this cartoon use the original broadcast masters, DVD releases don't always do so, which causes various issues.

I know.

Fun fact: I did some custom TF-related video work for Jay Bacal about 25 years ago. A couple things for Wizard/Toyfare at the time as well.
I was still hobby at the time, so I accepted swag as payment. I wish I still had the emails from Bacal, good conversations on TF. That was also about the time that TPTB started to notice my hobby work, but it was still several years before I'd accept pro work.

Unicron (Studio Cell 01) is on my desk.

If you're not in a hurry, I can pull some tapes when my current work is done in about a month or so. I know where all of those are.

[Sunstreak22]

Fun fact: I did some custom TF-related video work for Jay Bacal about 25 years ago. A couple things for Wizard/Toyfare at the time as well. I was still hobby at the time, so I accepted swag as payment. I wish I still had the emails from Bacal, good conversations on TF.

That must have been an interesting experience. Since Jay Bacal worked as creative director on this cartoon, he probably has interesting memories about Transformers.

Unicron (Studio Cell 01) is on my desk.

Cool, that figure is probably the most accurate version of Unicron from the original cartoon/movie.

If you're not in a hurry, I can pull some tapes when my current work is done in about a month or so. I know where all of those are.

No problem, I’m not in a hurry. This cartoon has many issues that I can work on over the next month in my free time.

[lordsmurf]

Fun fact: I did some custom TF-related video work for Jay Bacal about 25 years ago. A couple things for Wizard/Toyfare at the time as well. I was still hobby at the time, so I accepted swag as payment. I wish I still had the emails from Bacal, good conversations on TF.

That must have been an interesting experience. Since Jay Bacal worked as creative director on this cartoon, he probably has interesting memories about Transformers.

It was! I don't think I ever figured out where/how he found me, but it may have been Wizard. I just remember being in awe.

You have to remember that this was a VHS pre-DVD era. Back then, people involved with shows didn't even have their own copies of said shows to share with their kids ("this is what daddy did!"). In years long gone, a lot of hobbyists tried to help actors/writers/etc acquire copies (or snippets) of shows they were in. It was almost always to show kids or grandkids what they did, sometimes for their CV/resume. A lot of TF episode were unreleased back then.

Beyond that, I don't remember much. It was an instance of "wow, I'll remember this forever!" and then 25-30 later, I have no idea.

No idea what the payment/swag was either. Maybe a G1 toy? No Paypal then either, he mailed me something. Hmm....

Hmm....mail. Return address...

OMG! I still have his contact info, and it's still current!!!

You know, I'm not aware of anybody ever interviewing him. And we saw his name on literally every episode. Hmmmm....

[Sunstreak22]

You have to remember that this was a VHS pre-DVD era. Back then, people involved with shows didn't even have their own copies of said shows to share with their kids ("this is what daddy did!").

Now in the digital era, everything is easier. Analog media also have the drawback that, if stored in poor conditions, they can degrade or may get lost. I hope Hasbro still has the original film reels of the G1 episodes. They should hire a company to perform 2K/4K scans for archival purposes, remaster the episodes, and finally release the series on Blu-ray. We live in a time where HD is the standard, yet they’re still using SD masters.

You know, I'm not aware of anybody ever interviewing him. And we saw his name on literally every episode. Hmmmm....

I don't remember any interviews with him either. I tried to find something on the internet but I couldn't find anything. It's quite strange that no one has interviewed him after all these years.

[lordsmurf]

It's my understanding that
- The film production masters of G1 were lost in the early 90s.
- Rhino used the tape broadcast masters for the late 90s VHS releases.
- Rhino the used film pre-production masters for the early 2000s Rhino DVDs. Those masters had many errors, and then the Rhino S1-S2 DVDs also sucked for adding stupid sound effects (S3 was mostly fine).
- Shout later took the film pre-production masters, and swapped out bad/wrong scenes from the broadcast masters. Of course, you can tell immediately when the sources change, because of the upscale algorithms available in the late 2000s (and those were just released in the mid 2010s).

So that's where we are at.

Even setting aside the master source issues, some of the digital processing simply sucked. And that's what you're want to fix here. The first episode, in that first scene on Cybertron, has always been a pet peeve of mine. The first thing you see about Transformers is an aliased mess. Lots of younger fans, even older fans that don't know any better, have let that cloud their idea of how quality the show was from an animation stance (story can be debated). They can't shake the quality they see in that first scene. This is actually why I tell people to watch other episodes as their intro to the show.

[SaurusX]

Even setting aside the master source issues, some of the digital processing simply sucked. And that's what you're want to fix here. The first episode, in that first scene on Cybertron, has always been a pet peeve of mine. The first thing you see about Transformers is an aliased mess. Lots of younger fans, even older fans that don't know any better, have let that cloud their idea of how quality the show was from an animation stance (story can be debated). They can't shake the quality they see in that first scene. This is actually why I tell people to watch other episodes as their intro to the show.

You gave a perfect summary of The Transformers digital format releases. By doing a close comparison, you can see that Shout Factory also contributed to the mess despite making the effort to correct some animation mistakes. Their DVD video quality is definitely below Rhino's. On SF the strong reds, of which there are a lot, are aliased along the edges and there's noticeably more macroblocking over all. Of course, Rhino has the sporadic problem of interlaced chroma going over into adjacent frames. It's a difficult restoration project.

[Sunstreak22]

- The film production masters of G1 were lost in the early 90s.

Not necessarily, I don’t recall anyone involved in the creation or restoration of this cartoon mentioning that the film production masters were lost.

- Rhino used the tape broadcast masters for the late 90s VHS releases.

Most likely, yes.

- Rhino the used film pre-production masters for the early 2000s Rhino DVDs. Those masters had many errors, and then the Rhino S1-S2 DVDs also sucked for adding stupid sound effects (S3 was mostly fine).

Rhino used film pre-production masters, but not for all episodes (some are entirely from analog sources, and there may be some from the film production masters). Someone described many errors on this page: https://web.archive.org/web/20190124085714/http://cartoonreviewsite.com/changes/tf.htm. According to this, some episodes have no changes compared to the broadcast masters, but certain segments might have been replaced with analog sources. Overall, the number of errors varies by episode. Their release has two audio tracks (5.1 and stereo). In some episodes, the stereo track does not have these unnecessary added sound effects (for example, the first three episodes have the original audio, but only in stereo).

- Shout later took the film pre-production masters, and swapped out bad/wrong scenes from the broadcast masters.

Someone who worked on the Shout release mentioned that Hasbro provided them with the same masters that Rhino used for their release. They replaced the errors using original broadcast one-inch masters (confirmation of this can be found here: https://tfwiki.net/wiki/Shout!_Studios#cite_ref-9). Shout corrected more than Rhino did, but not everything, and they also made a few additional errors.

Even setting aside the master source issues, some of the digital processing simply sucked. And that's what you're want to fix here.

Yes, I would like to improve the video quality and fix other issues that the episodes should not contain.

[Sunstreak22]

Of course, Rhino has the sporadic problem of interlaced chroma going over into adjacent frames.

S02E20 frame 276 (Rhino):

[Attachment 81888 - Click to enlarge]

Do you mean something like this?

In the Rhino release, starting from episode 9, progressive frames with soft pulldown begin to appear. Even for progressive frames, TFM can incorrectly detect combed frames, although setting the mode to 4 fixes that issue. If I'm not mistaken, when the progressive scan changes to an interlaced scan, chroma blending occurs (e.g., S02E20 frame 276). The same frame in the Shout release doesn’t have this problem, probably because all the frames are interlaced.

The same frame from the Shout! Factory release:

[Attachment 81889 - Click to enlarge]

The Shout! Factory release seems better for some interlaced segments, while Rhino appears to be better for progressive segments because it avoids issues with incorrect field matching. I would like to determine at which frame numer the switch from progressive to interlaced occurs. Is there a way to identify which frame is progressive and which is interlaced?

[SaurusX]

S02E20 frame 276 (Rhino):

[Attachment 81888 - Click to enlarge]

Do you mean something like this?

Correct. Rhino's encoding method seems to have introduced this effect, though it can be undone. See the doom9 thread here.

The Shout! Factory release seems better for some interlaced segments, while Rhino appears to be better for progressive segments because it avoids issues with incorrect field matching. I would like to determine at which frame numer the switch from progressive to interlaced occurs. Is there a way to identify which frame is progressive and which is interlaced?

DGIndex identifies frames marked as progressive or interlaced as it indexes the video. TFM then uses the "flag" parameter to determine whether field matching and comb detection should be done. "Flag" is already set to "4" by default, which sounds like it should be ideal, but leaves us exactly where we already are.