What is the best way to deinterlace and upscale films from DVDs using the Dione and Proteus models

[dangromann]

Thank you so much for all the replies! I appreciate your taking the time to post them. They are potentially very useful, although they also do raise new questions for me. I will post the most pressing ones below.

It depends on the footage you process, but most films are produced in 24p (23.976fps progressive).
Later DVDs are often recorded in 24p, but early DVDs may be telecine and recorded in 30i (29.97fps interlaced).
For DVDs recorded in 30i, Inverse Telecine(30i → 24p) is required.

How would I know if the particular footage needs Inverse Telecine (IVTC) or not?

However, the current TVAI does not have an Inverse Telecine function.

Interestingly, among the films I have upscaled so far (always using the Proteus model), there have been some cases where the source video (a specific example: a 1948 film ripped from a 720x480 DVD) was 29.97fps (at least judging by what VLC is showing me when I play the file; MediaInfo is not showing any fps information at all in this case) and after processing with Proteus, the resulting video is 23.976fps. Would that be because IVTC happened somehow? I was actually quite amazed to see that change (which did not effect the length of the film). Or is it that the VLC information regarding the source file may be incorrect? Could it be the case of a “later DVD recorded in 24p”?

have you tried the simpler option of Deinterlacing and upscaling using Dione: DV (or TV) in one shot as a single step?

Yes, this is pretty much option 2 I mentioned in my message: " 2. Deinterlace, upscale and convert pixels to square ones using Dione" - all in one single step. The results seem good to me. The resulting video is 59.940 fps.

Taking this into account (the seemingly good result), and contrasting it with this statement:

For DVDs recorded in 30i, Inverse Telecine(30i → 24p) is required.

would that imply my footage did not really need IVTC? If so, this brings me back to the question of how would I know, before processing, that the footage needs IVTC or not.

A further question: if I end up with a 59.940 fps video (movie) after the processing, is there anything wrong with it? I mean, would it have been better to have converted the footage to 24fps before upscaling (to also end up with a 24 fps processed video)? From what I’ve read on the web, people would say there is no ideal frames per second rate, it all depends on what I need, and they would say 24fps gives a “movie” feel to the video, while doubling the fps when deinterlacing (to 59.940 fps in case of a 29.97 fps source video) is highly recommended and ensures smoother motion. The Dione model does exactly that (doubles the fps when deinterlacing), so if the resulting video looks good, am I done or should I worry it’s not 24 fps?

And the last question for now: regarding the first method I mentioned - deinterlace the 720x480 video (ripped from a DVD) and convert it to square pixels using Dione (without upscaling), then, as a separate process, upscale it to FullHD using Proteus - is there any reason why this would be not recommendable? (like that for example it’s not advisable to process the same video twice using different models or something - I’m asking because I don’t know if there may be any such rules).

Thank you in advance for any replies .

[Akila]

@dangromann

And the last question for now: regarding the first method I mentioned - deinterlace the 720x480 video (ripped from a DVD) and convert it to square pixels using Dione (without upscaling), then, as a separate process, upscale it to FullHD using Proteus - is there any reason why this would be not recommendable? (like that for example it’s not advisable to process the same video twice using different models or something - I’m asking because I don’t know if there may be any such rules).

I can’t answer you on the other questions you got as I got zero knowledge on Telecine.
but on this one, I actually recommend if you are planning to go that route of de-interlace your video 1st as a separate process, to de-interlace with an external program 1st using the QTGMC + bob method (not sure how it plays well with Telecine, maybe the telecine people here can comment) and not through TVAI. export it in a “lossless” video (see screen shots)
for this example I’ll be using “Hybrid” which is pretty popular tool and it is a freeware.

De-interlacing filter settings:

Lossless Settings (pick one of the two lossless encoding options x264 or FFV1, both are fine)
H.264 / x264

OR (in this one, if you choose, make sure to output as “avi”, not “mp4”, under “Base” tab, bottom left)
FFV1

to select what encoding format you use (e.g. lossless x264 or FFV1) for the two lossless options above.

NOTE: bear in mind, encoding in “lossless” would generate large file size, so make sure you have enough storage available.

once you de-interlaced your video, you can feed TVAI with your de-interlaced video and use your filter/preset setting of choice.

P.S. I have noticed personally that “Original Pixel” settings in TVAI generated sharper and slightly more detailed/quality results then selecting “Square Pixel” on my 1998 Video8 4:3 576i camcorder movies

you can test both for yourself and preview both on TVAI and if you wish you could also use a compare tool to check the results to see for yourself if you notice any difference between the two (that would save you tons of time). or just playback the 2s / 5s preview you generate from both one right after the other to see if you spot any difference in quality.

Video-Compare - Video compare tool
Topaz Video AI v3.0: Working with multiple video files - YouTube - how to use preview mode

[TicoRodriguez]

How would I know if the particular footage needs Inverse Telecine (IVTC) or not?

There is a detailed explanation on wikipedia.
(“Telecine” and “pull-down” are originally two different words, but are generally used interchangeably.)
The process diagram will help you get an idea of how the conversion between 24p and 30i is done.

To determine whether or not a video is a pulled-down video, feed the video frame by frame in the video editor.
If it looks like “A,” “B,” “B+C,” “C+D,” or “D,” then it is most likely video being pull-down.

Or is it that the VLC information regarding the source file may be incorrect? Could it be the case of a “later DVD recorded in 24p”?

Please read “Soft and hard telecine” on the “Telecine” page.
Older or lower quality DVDs may be recorded in hard telecine (30i), while newer or better quality DVDs are recorded in soft telecine (24p → 30i decoding).
Some players may recognize soft telecine as 30i.
I tried to display the information in Mediainfo and VLC for videos encoded with Soft Telecine.

A further question: if I end up with a 59.940 fps video (movie) after the processing, is there anything wrong with it?

The image quality is reduced because the original resolution is low and the amount of information is low.
720x480 30i (29.97 fps) is like 720x240 at 59.94 fps.
It is smooth as movement, but the amount of information as a picture is half the resolution.
For telecine video, inverse telecine is performed to 720x480 24p to obtain the appropriate amount of information.

Telecine

Three-two pull down

[Carlito]

Maybe this also help… Interlacing patterns

[ssbroly]

Hi, thanks 1000 times for the lossless encoding method! For a long time I used the Prores thinking it was the best, I tested your method with the H264 in lossless, there is a big difference in quality! I also work with camcorder videos from 98 to 720x576. On the other hand, when I used the Prores, with QTGMC, I did not put everything in slower, the more we increase the “final temporal smoothing”, the more details we lose in the movements.

[dangromann]

Thank you SO MUCH to akilaspam and TicoRodriguez for your detailed replies, and also to Carlito for the link. I greatly appreciate all the info and tips!

The information (which I’ll still have to ‘digest’ fully) already makes the whole picture much clearer. And the lossless method of de-interlacing in Hybrid sounds just like what I need. I’ve used Hybrid before (a little bit), so I’m slightly familiar with it.

I will have to do my homework now. I will get back with feedback about the results or more questions.

[Akila]

@ssbroly

Hi, thanks 1000 times for the lossless encoding method! For a long time I used the Prores thinking it was the best, I tested your method with the H264 in lossless, there is a big difference in quality! I also work with camcorder videos from 98 to 720x576. On the other hand, when I used the Prores, with QTGMC, I did not put everything in slower, the more we increase the “final temporal smoothing”, the more details we lose in the movements.

that is why i stopped using ProRes, I noticed it has weird quality issues on my Video8 processing in some cases (i posted a whole new post about this issue here in the forum).
now I only use the lossless x264 and the lossless FFV1 (started to like that one). I personally couldn’t tell the difference in quality between the two, to me they look the same.
once you de-interlace your videos, check out @TomaszW presets for upscaling.
I personally found two that works great for my Video8 files.

TVEAI 3x.. Presets V1, V2

try the “8mm_to_1080” or the “8HUltraHD” presets, they result in pretty sharp and I found them both being very good for upscaling my lower quality videos (Video 8 files). I also found the VHS to HD preset yields blurry image.

[TicoRodriguez]

Was QTGMC able to do inverse telecine (30i → 24p)?
I thought it was used to convert 30i to 30p or 60p.

[ssbroly]

Indeed the prores is to be banned in fact I haven’t tested with FFV1 yet, I’ll test when I have time, are you talking about hybrid or vai presets? PS: the final temporal smoothing loses less detail in H264 Lossless than in Prores, so it’s great! but with H264, the files are only 65 MB / s and in dark passages, it is like kind of spots.

[Carlito]

Which Prores profile are you using in Hybrid, LQ or HQ ? LQ compresses more and for that it removes even more noise and details…

[Carlito]

QTGMC is a deinterlacer, not an inverse telecine tool.

Best way to detelecine in Avisynth & a few examples of scripts I've tried.

[TicoRodriguez]

I knew I couldn’t do inverse telecine in QTGMC.
I think that inverse telecine is the best way to get the best image quality for a telecine (pull-down) video, so I am uncomfortable with the name QTGMC being mentioned.

[ssbroly]

I was using prores hq with hybrid and vai, but lossless h264 in hybrid followed by a vai model gives much better results. Afterwards with hybrid I always code in prores hq because I don’t know if h264 with the crf at 0 has less loss than prores. I am using version 2.6.4

[Carlito]

Same feeling here…

[Carlito]

CRF at 0 is lossless… Encode/H.264 – FFmpeg
However I don’t know if VEAI respetec the correct parameters for a true H264 lossless encoding…
I have to do extensive comparations between H264 and FFV1 to see if there are significant differences in quality between the two. @Akila mentioned before in this same topic that he can’t spot any difference between the two.
One think I’ve checked in the short tests I’ve done is that FFV1 encoding speed is faster than H264… 4minutes for a 42minutes source. If one are processing long sources only that difference in encoding speed is to be taken in account.

[Akila]

@TicoRodriguez

Was QTGMC able to do inverse telecine (30i → 24p)?
I thought it was used to convert 30i to 30p or 60p.

like I mentioned before, i got zero knowledge when it comes to Telecine, so I have no idea if it does, I only work with interlaced videos that were shot in it’s original fps as it was stored.

@ssbroly

Indeed the prores is to be banned in fact I haven’t tested with FFV1 yet, I’ll test when I have time, are you talking about hybrid or vai presets? PS: the final temporal smoothing loses less detail in H264 Lossless than in Prores, so it’s great! but with H264, the files are only 65 MB / s and in dark passages, it is like kind of spots.

if you mean FFV1, I am talking about Hybrid. you can see my screen shot earlier in my post in this thread. FFV1 is basically ffmpeg forking of h264 to my understanding, so i am not expecting any quality difference to be honest, but performance might be different, i think it’s a bit quicker then h264.

[ssbroly]

Yes I’m talking about hybrid, I’m still testing the H264 at the moment. But you say that it is lossless and that the files are therefore large, but they are 95 MB/s in AVI H264, so it’s not so big I find

[TicoRodriguez]

For TVAI’s FFMPEG, x264 or x265 cannot be used, so h264 such as nvenc or qsv must be used.

I looked up the options for h264_nvenc.
As for lossless, it seems to be specified in preset or tune (I haven’t tried it).

"C:\Program Files\Topaz Labs LLC\Topaz Video AI\ffmpeg.exe" -h encoder=h264_nvenc

Encoder h264_nvenc [NVIDIA NVENC H.264 encoder]:
    General capabilities: dr1 delay hardware 
    Threading capabilities: none
    Supported hardware devices: cuda cuda d3d11va d3d11va 
    Supported pixel formats: yuv420p nv12 p010le yuv444p p016le yuv444p16le bgr0 bgra rgb0 rgba x2rgb10le x2bgr10le gbrp gbrp16le cuda d3d11
h264_nvenc AVOptions:
  -preset            <int>        E..V....... Set the encoding preset (from 0 to 18) (default p4)
     lossless        10           E..V....... 
     losslesshp      11           E..V....... 
  -tune              <int>        E..V....... Set the encoding tuning info (from 1 to 4) (default hq)
     lossless        4            E..V....... Lossless

But the problem is that h264_nvenc is limited to 8-bit output.
TVAI 2.x.x has 8bit internal processing so there is little loss, but TVAI 3.x.x has 16bit internal processing so it would be a waste to output at 8bit.

[Carlito]

I need to test NVENC to see if there is any speed gains and if at least NVENC lossless quality is on par with Cpu encoding.
Since I’m processing 8bit sources I don’t know if I gain anything in outputting to 16bit…it doesn’t hurt to try anyway.
Thanks for your info

[djtlmed]

Thanks! Thought I’d share that ChatGPT is a good sidekick for all these procedures.

https://chat.openai.com/share/c60b9c3c-a669-44ae-901a-3ebe92c3f8b2