A couple of weeks back I started bouncing around the forum in the various topics asking for advice on my first attempt at a restoration. Thank you to everyone who offered advice and experienced insight.
If anyone would care to take a gander...here are 2 links...
The first one offers some split screen before and after shots of the project:
The second one is a link to the completed film (and also generically explains what I did):
Constructive criticism is always welcomed. This was my first attempt at such a project, and I look forward to learning more and discovering new challenges ahead.
+ Reply to Thread
Results 1 to 30 of 35
Your results look extremely good. I have nothing to suggest. Were you using some of VideoFred's AVISynth scripts, or did you use something completely different? I was particularly impressed by the amount of detail you were able to coax out of the old film.
Last edited by johnmeyer; 26th May 2020 at 11:47. Reason: typo
I did not use any scripts in particular. I used Hybrid to do the deinterlacing (Vapoursynth), and then used the Avisynth section within Hybrid to stabilize, degrain, and adjust the brightness and contrast. I just kinda played with all the parameters until I got something I liked. A lot of trial and error.
I didn't know how much of an effort in futility this would be, since my source was a (seemingly) crappily prepared DVD. Glad you liked he results.
Why 50fps? The frame interpolation looks godawful horrible. Every interpolated frame during movement shows blending. You didn't notice that?
I'd have slowed it to film speed so the video would be at the correct speed and audio at the correct pitch.
You had me until you mentioned the audio track. At that point, I knew not to take you too seriously. The audio was rendered completely independent of the video and audio is my specialty. It's speed and pitch are perfect.
Perhaps I am getting defensive too harshly (in spite of that bogus audio claim)....
I asked for constructive critical analysis, and I meant it. It's the only way to learn.
Here are 6 frames from the final rendering in a shot that has more movement than anywhere else in the film. Which are the interpolated ones, and where is the blending?
[Attachment 53527 - Click to enlarge]
[Attachment 53528 - Click to enlarge]
[Attachment 53529 - Click to enlarge]
[Attachment 53530 - Click to enlarge]
[Attachment 53531 - Click to enlarge]
[Attachment 53532 - Click to enlarge]
I corrected the pitch/speed....it was a tad sharp. Which makes sense as per your description, John.
However, I owe you an apology about the interpolation accusation. Apparently, I was only looking at the source. Picture below. But why is it like that? How was it 'deinterlaced'? Do you know? Do you have a sample from the original DVD?
I was also surprised at the gain in clarity and sharpness. Now I think the source was made significantly worse so, in comparison, the restoration looks better.
Last edited by manono; 26th May 2020 at 15:20.
I deinterlaced it in Hybrid using Vaporsynth / QTGMC, Top field first.
As for the pitch being perfect....I went by the music in the opening and closing credits.
It was just a tad sharp and I corrected the entire audio track accordingly. Music doesn't lie!
However...if the musicians were out of tune when they recorded it originally....well, at that point I would have to give up!
Also, the whites are often blown out. Even the blacks are crushed in places, especially during the fades and the black columns on the sides (why add the black columns?). If you were to put on:
You'd get something like the below picture (green marks blown out whites, red the crushed blacks). Apparently, you added the black bars before messing with the brightness and contrast. The text doesn't much matter.
I can answer some of the questions. For pitch, if there is music, you can assume the conductor tuned the orchestra and you can correct your pitch based on comparing the notes on the soundtrack to reference tones. It's pretty easy to do because the 25/24, while close to a semitone, is not going to produce notes that are perfectly "on-key."
As for motion interpolation, after manono made that comment I went back and looked. I can state with absolute certainty that no motion interpolation was used. How can I say that? Because a horizontal pan across a picket fence is the most difficult torture test there is for testing motion interpolation. Those vertical members will break and warp no matter what algorithms you use, and no matter what tuning you use. I saw zero morphing or distortions.
The real problem is the blending that was in the original but, if you look at the OP's "after", his software did a rather brilliant job of extracting some semblance of the original un-blended frame. Very cool stuff, IMHO.
Sound Forge as my WAV editor, and Pro Tools, and/or Sonar Professional (depending on the project) for recording along with the Mercury VST Bundle from Waves.
Last edited by Joey Bagodonuts; 26th May 2020 at 16:58.
It's pretty easy to do because the 25/24, while close to a semitone, is not going to produce notes that are perfectly "on-key."
The real problem is the blending that was in the original...
DGIndex. Scroll to the bike rider and use the [ and ] buttons to isolate 10 seconds or less. Then File->Save project and demux video. Please make the resulting M2V video file available.
With your perfect pitch, can you tell if pitch corrected audio I've created from PAL sources might be accurate or not?
And I will post the bicycle shot tommorow....
Last edited by Joey Bagodonuts; 26th May 2020 at 17:56.
The 2 pics I took aren't any of the frame pictures you included. But those aren't blends in that first pic. Blurring, I think. Maybe YouTube's reencoding of the restoration demo did it.
Okay, I'll link first to a song video in one of my channels. The second link will be to a PAL version similar to what I started with. Although it was an NTSC DVD I worked on, it was made from a PAL master and the 'effective' framerate of that NTSC DVD was 25fps. So, the song went from its original 24fps film speed to PAL's 25fps, and in my upload back to 23.976fps.
Mine was really made at 23.976fps but I doubt you can tell the difference between 23.976fps and 24fps. Can you?
I HAVE to assume that the intended original key, especially based on the instrumentation, is C.
The 1st of the 2 clips is considerably more flat (ie: slower) than the first clip. I'd say by about 1/2 of a semitone (ie: a quarter-tone).
I would hazard a guess based on these clips then that they are BOTH running slow, but the 2nd one is closer to where it needs to be.
You say slow but the 2nd one is 25fps and the first one runs at film speed. So, you're not using 'slow' to mean film speed, but the audio?
The first one is a quarter-tone flat and the second one also flat, but less so?
Okay, thanks for looking. Back onto the subject. Personally, I see no need to interpolate to 50fps. This is a restoration, or it's claimed to be. Now, if you're only experimenting with what the software can do, that's one thing. Then it might be labeled an experiment and not a restoration. That's how one learns - by experimentation and reading how others have approached the same thing. Good.
I think your software must work in RGB but it wasn't brought back to YV12 properly. That might explain the illegal values for the blacks and whites. But that's more pdr's and jagabo's specialty, I think. Not mine as all my work is done in AviSynth and, except when capturing VHS tapes, all in YV12.
I download an MP3 audio version of the two video clips. I then stretched one to match the other. The stretch ratio for 25/24 is 1.041666667. The stretch ratio to make the two audio tracks match is 1.049377865. These two do not match exactly. Since I am not an expert in film to PAL conversions, I don't know what else might have been done. So this may not have been done by simply speeding up the film version. (BTW, even adding a 1000/1001 correction doesn't make the two numbers agree.
But, the two sound very different to me, besides the other one playing noticeably faster. The other one sounds noticeably higher pitched to me. And I don't have perfect pitch and am in no way a musician.
By the 1920s, A440 was the standard in America. So unless you are looking to distinctively be different for whatever odd artistic purpose, it was 440. Especially when an equal temperament instrument is playing in the arrangement...ie: a piano...as the soundtrack in this movie had.
These movies were made cheaply and quickly, and music at that time was the last thing they were thinking about. There is no reason to think that anyone would put that kind of effort into writing and recording music at A432 or something similar.
Last edited by Joey Bagodonuts; 27th May 2020 at 06:49.
I would be EXTREMELY surprised if the original key was B, given the instrumentation of the music, the type of music it is, and the rarity of such music being in a difficult performance key for those particular instruments. I'd put hard money on the proper pitch of the track being in C.
Here is the Picket Fence shot directly from the DVD....
Finally...here are the 2 frames that resemble most closely the frame that monono referenced above with that funky picket fence issue. I don't see the same from the individual frames here...so I either botched up the final render, or it's a YouTube thing....
[Attachment 53539 - Click to enlarge]
[Attachment 53540 - Click to enlarge]
The original is 25p content; it's just "out of phase" . You can use TFM to match the fields .
It looks like the "original" comparison on YT used frame blending for the 25p=>50p conversion (ConvertFPS in avsiynth would cause blending, ChangeFPS would use duplicates; or the "default" setting in most NLE's would be frame blending too) . But likely that's what the problem is on youtube with the "original" side