Not converting Interlaced footage to Progressive if viewing on BD player?

[clashradio]

After months of research and a bunch of tests with various equipment to get the best possible capture from laserdiscs I’m baffled at the prospect of leaving the footage Interlaced if viewing on a BD player-only vs. uploading online. 99% of the forums say to deinterlace and IVTC. Yet only a few say to let the player do the deinterlacing.

I’m capturing at 480i YUV 10-bit 4:2:2 uncompressed. DVD recorder (passthrough) for 3D comb filter. YADIF for deinterlacing. Upscaling to 720p 23.97fps. So far good results but a few projects have more combing/ghosting on faster moving objects than others. I know with Interlaced footage it will never be as good as if the 3:2 pulldown was never applied. And I know QTGMC is a better method for deinterlacing than YADIF.

Questions:
1) Can a BD player deinterlace better than QTGMC?
-I would think a computer program would be better than a piece of hardware. I originally bought a video processor for the 3D filter, which is also capable of real-time IVTC & upscaling but learned that a program would do a better job with both.

2) If the player does the deinterlacing, what about IVTC? If leaving the footage Interlaced which is 29.97fps, what if the original source is 24fps?
-For the discs with footage shot on video at 29.97 I’m sure it’s not a problem.

3) Little confusing why Progressive 23.97fps isn’t actually 24fps? A movie shot on film at 24fps should equal Progressive video 24fps since there are no fields involved?


Thanks to usually_quiet for convincing me capture at the native frame rate & resolution of 480i/29.97. And to use a dvd-recorder for the 3D comb filter instead of a video processor. That was a hard pill to swallow as the Lumagen 2144’s comb filter is absolutely perfect. But the lowest output rez is 480p and the only interlace option is 1080. Not to mention the overscan issues the 2144 has with analog the inputs.
Also thanks to Cornucopia, JV Raines, jagabo

[jagabo]

1) Can a BD player deinterlace better than QTGMC?

No. But someday in the future there may be better hardware for deinterlacing. Any video you deinterlace now will have the deinterlacing quality locked in. But if you leave the video interlaced your video will look better on such hardware.

2) If the player does the deinterlacing, what about IVTC? If leaving the footage Interlaced which is 29.97fps, what if the original source is 24fps?

Some players have film detection and will IVTC when possible. But progressive encoding is superior to interlaced encoding. So IVTC before encoding when possible.

[B]3) Little confusing why Progressive 23.97fps isn’t actually 24fps?

For technical reasons, NTSC color TV was set at 59.94 fields per second, aka, 29.97 frames per second. Film has traditionally been slowed from 24 fps to 23.976 fps when telecined to 59.94 fields per second.

[clashradio]

1) "any video you deinterlace now will have the deinterlacing quality locked in". That's what I'm afraid of. Yadif is better than the built-in deinterlacers of Sony Vegas Pro, but not as good as QTGMC. I wonder how Yadif compairs to the BD player's deinterlacing.

2) In past posts you've mentioned that IVTC gives better quality results than deinterlacing. I don't understand how as IVTC is just removes the added frames where as deinterlacing gets rid of the combing/ghosting effect.

3) When film sourced footage is transferred to HD and put to BD it's usually in Progressive, but I do have a few BD's in 1080i. If in Progressive, shouldn't the frame rate be 24fps, not 23.97? since there are no fields to deal with? If I understand correctly for HD; one frame of film equals one frame of video where as SD one frame of film is basically two fields of video?

4) I forgot to ask this in my original post: When viewing laserdiscs from the LD player on my CRT tv via composite, I don't see any interlacing going on. Is it because I'm viewing on a CRT vs. HD monitor? I've got player flags/interlace/progressive on the brain.

5) Some people have done IVTC manually. Does this mean they've taken a 2hr movie and manually cut the duplicate frames frame by frame?

6) In MediaInfo, how do I tell what the scan order is? I can't see it on my example?

7) What would happen if I deinterlaced a video-sourced laserdisc vs. film-sourced?

[jagabo]

1)I wonder how Yadif compairs to the BD player's deinterlacing.

Probably better than some, not as good as others.

Regarding QTGMC: I suspect there may never be hardware better than QTGMC. The industry is more likely to switch to pure progressive first.

2) In past posts you've mentioned that IVTC gives better quality results than deinterlacing. I don't understand how as IVTC is just removes the added frames where as deinterlacing gets rid of the combing/ghosting effect.

IVTC first matches fields to reproduce the original film frames. It then removes the one duplicate out of every five frames (for 23.976 to 29.97 telecined sources). The field matching is usually very accurate (few errors) and gives better results than deinterlacing, even with QTMGC. There are some exceptions: very bad VHS captures (especially with lossy codecs), poor interlaced MPEG encoded videos, etc.

BD...If in Progressive, shouldn't the frame rate be 24fps, not 23.97?

Sometimes the industry just goes with what it's always done.

4) I forgot to ask this in my original post: When viewing laserdiscs from the LD player on my CRT tv via composite, I don't see any interlacing going on. Is it because I'm viewing on a CRT vs. HD monitor?

Yes. CRT TVs are inherently interlaced devices. You never see both fields at the same time on a CRT. Watch this video to see high speed video of an interlaced TV. You can see the individual scan lines being drawn. Note how most of the screen is black at any one time. This is why CRT TVs flickered some much with some material.

https://www.youtube.com/watch?v=3BJU2drrtCM

5) Some people have done IVTC manually. Does this mean they've taken a 2hr movie and manually cut the duplicate frames frame by frame?

It usually means they ran a filter that reconstructs the film frames assuming a fixed pattern. If there is a break in the pattern they have to work in sections, using the right pattern for each section. IVTC with TFM().TDecimate() can automatically adjust to such pattern changes. Cadence changes can be caused by the video being edited after being telecined (eg, commercials inserted, scenes cut for running time, frames dropped during capture, etc.)

6) In MediaInfo, how do I tell what the scan order is? I can't see it on my example?

MediaInfo doesn't seem to realize that video is interlaced. To check such videos I use the Bob() filter and watch motion. If you see back and forth motion you have the wrong field order (AviSynth assumes BFF if not otherwise specified). MediaInfo usually reports the field order if it realizes the video is interlaced. But don't blindly assume it's correct. Sometimes video is encoded with the wrong field order. Sometimes progressive frames are encoded interlaced (very often with PAL captures) and don't need to be deinterlaced.

7) What would happen if I deinterlaced a video-sourced laserdisc vs. film-sourced?

If you use QTGMC and don't decimate you will end up with a 59.94 fps video with lots of (very near) duplicate frames. You'll need a higher bitrate to encode the video and playback requires more power (some device can't handle 1920x1080 at 59.94 fps, for example). QTGMC sometimes make mistakes and creates ghost artifacts -- especially with animated material. Or moire artifacts when the video has closely spaced horizontal lines. When you display a 1920x1080p59.94 video on a display who's refresh is not 59.94Hz (a 50 Hz PAL TV, fore example) it may be jerkier than when playing a 23.976 fps IVTC instead. Sometimes people use QTGMC in progressive mode after IVTC -- to take advantage of its other cleanup abilities.

[clashradio]

Thanks jagabo. I apologize if these questions have already been answered in some past post. That Youtube vid was a mind blower. I know I still need to understand IVTC and deinterlacing better...I'm trying lol.

Pretty sweet how qtgmc is better than the set-top players deinterlace mode.

So you're saying a computer's bd player program deinterlaces when viewing discs on a PC?

[jagabo]

So you're saying a computer's bd player program deinterlaces when viewing discs on a PC?

Yes, if the video is interlaced. Progressive sources are just frame duplicated from the source frame rate to the display's refresh rate.

[jagabo]

I still need to understand IVTC and deinterlacing better...

Consider four film frames, A, B, C, and D. To be broadcast on interlaced TV those 4 frames must become 10 fields. This is done by displaying the frames for either 2 fields or 3 fields, alternating between the two. On average each film frame is displayed for 2.5 fields. 24 fps * 2.5 = 60 fields per second. Let's use capital letters for top fields, lower case letters for bottom fields. The TV always alternates between top and bottom fields. So using 2:3 pulldown we get the following:

A a B b B c C d D d

Note film frame A is repeated twice, B is repeated three times, C is repeated twice, D is repeated three times... And the display always alternates between top and bottom fields. When you digitize that sequence pairs of fields are joined together into interlace frames:

Aa Bb Bc Cd Dd

Notice how the first two frames and the last frame each contain both fields from one of the original film frames. But the third and forth frames contain fields from two different film frames. And the two fields for film frame C are split between those two video frames. You can perform a manual IVTC by first splitting the frames back into fields, selecting which fields you want to keep (8 out of the 10), and weaving them back together into frames:

                       # video frames: Aa  Bb  Bc  Cd  Dd
SeparateFields()                     # A a B b B c C d D d
SelectEvery(10, 0,1, 2,3, 5,6, 8,9)  # A a B b   c C   D d
Weave()                              # Aa  Bb    cC    Dd

We have restored each of the original film frames and there are no duplicates. Note that SelectEvery(N, ....) means: for every N frames (really fields here) keep the following frames (discard the rest).

But what happens if the video was edited and the first video frame is missing:

Bb Bc Cd Dd Ee

Ee is from the next group of video frames. If you run the same code on this you get:

SeparateFields()                     # B b B c C d D d E e
SelectEvery(10, 0,1, 2,3, 5,6, 8,9)  # B b B c   d D   E e
Weave()                              # Bb  Bc    dD    Ee

The result contains frames with fields from two different film frames. And only one field of film frame D appears in the sequence. I'm sure you can see how SelectEvery() needs to be changed to handle this sequence.

TFM() can adapt to changes like this. Basically it starts with one field of a frame then looks at the field before and the field after. It combines it with whichever field leads to the least comb artifacts. So lets say it starts with the first bottom field:

video frames: Aa Bb Bc Cd Dd
start with first bottom field a:  Aa Bb Bc Cd Dd
A and a are paired together: Aa
move to the next bottom field b:  Aa Bb Bc Cd Dd
b is paired with either B:  Aa Bb
move to the next bottom field c:  Aa Bb Bc Cd Dd
cis paired with C after it. Aa Bb cC 
move to the next bottom field d:  Aa Bb Bc Cd Dd
d is paried with D after it: Aa Bb cC dD
finally the last bottom field d:  Aa Bb Bc Cd Dd E...
d is paired with the D before: Aa Bb cC dD Dd

We now have five film frames but one of them is a duplicate. After TFM() we use TDecimate() to locate and removes the duplicate:

Aa Bb cC dD Dd --> Aa Bb cC dD

In the end we have the four film frames from the five video frames.

Sometimes edits or dropped video frames cause orphaned fields -- ie, only one field of the original film frame appears in the video. In those cases you can't fully restore the original film frame so TFM() will deinterlace the frame instead.

[clashradio]

Hi jababo. I was wondering if you could do a QTGMC test with this 3 second clip? It still looks pretty ghostly when using Yadif.

[jagabo]

That's a tough clip. The fast cuts make it hard to see what's going on when watched in real time. It's basically interlaced video (59.94 fields per second). but the center portion, with the guitarist jumping in the air was slowed down by duplicating and shifting fields (ie, converting top fields to bottom fields or vice versa). This screws up the deinterlacing and causes the picture to bounce up and down. But here's what I got with Yadif on the left, QTGMC on the right.

AviSource("Paul's arm.avi") 
Crop(12,6,-4,-0) # 704x480
AssumeTFF()
ConvertToYV12(interlaced=true)
StacKHorizontal(Yadif(mode=1), QTGMC())

[clashradio]

Ya it's hard to tell due to the pixilation. Thank you though for the tests.

[jagabo]

Look closely at the horizontal edges of the drums at the start of the clip. They're all pretty clean in the QTGMC side. The Yadif side has lots of aliasing artifacts. That is typical of the differences between the two methods.

[clashradio]

Look closely at the horizontal edges of the drums at the start of the clip. They're all pretty clean in the QTGMC side. The Yadif side has lots of aliasing artifacts. That is typical of the differences between the two methods.

I dropped your clip in my Vegas Timeline to analyze frame-by-frame.......wow!! QTGMC blows Yadif out of the water. I don't even want to use Yadi anymore now

I have to admit I thought IVTC was only for film-sourced footage? I didn't know it also worked for video sourced stuff like this KISS clip. I thought the process was only to take the 29.97fps back to 23.97fps?

[jagabo]

Yadif and QTGMC deinterlace, not inverse telecine. IVTC does not work for this clip (aside from a few frames in the slowed down section).

The main disadvantage of QTGMC vs. Yadif is speed. QTGMC is very slow. For example, on my computer, single threaded, a 720x480 DVD source is processed by Yadif at about 500 frames per second. QTGMC at its default settings runs about 13 frames per second.

[clashradio]

I don't mind the slow render speeds if QTGMC does a better job in the long run. Like you said any sub-par deinterlacing I do now is stuck at that level of quality on the BD.

It would be nice if AVIsynth was more user friendly. I'm assuming you have to have that to use QTGMC?

[jagabo]

I believe there's also a version for VaporSynth but that's no more user friendly than AviSynth.

It's possible to use QTGMC with MeGui (GUI front end for AviSynth and encoding) but you still have to track down and install all the filters yourself.

[clashradio]

Hopefully this isn't too elementary of a question but... So no matter what the original source on the ntsc LD is; Hollywood movie @24fps or a music video compilation that might be 29.97, the captured footage definitely could be improved by QTGMC? and if you're going through all the hassle of capturing it probably doesn't make sense to even use Yadif, it's QTGMC or nothing right? When using Vegas Yadif does seem to be a slight improvement over the built-in options of Blend/Interpolate fields.

As far as IVTC goes, does it matter if the original source is film or video? Just wondering because of the different frame rates.

And again, you're saying if deinterlaced by QTGMC, you may not need to do IVTC?

[manono]

Hopefully this isn't too elementary of a question but... So no matter what the original source on the ntsc LD is; Hollywood movie @24fps or a music video compilation that might be 29.97, the captured footage definitely could be improved by QTGMC?

For video sources, yes it's probably best to use QTGMC. For film sources, almost always no. If the source is a DVD or even a laser disc, you perform an IVTC. You want to get back the original untouched frames. If you're using QTGMC purely for its cleaning properties, in my opinion better (and usually faster) would be to use dedicated cleaning filters.

...it's QTGMC or nothing right?

I use Yadif only for testing, because it's fast. I don't use it for either deinterlacing or for bobbing.

As far as IVTC goes, does it matter if the original source is film or video?

Of course it matters. An IVTC is for film sources. Using it on video will ruin it.

And again, you're saying if deinterlaced by QTGMC, you may not need to do IVTC?

QTGMC bobs a 29.97fps source to 59.94fps. To return a film source to its native frame rate after bobbing and do away with all the duplicate frames created, you should use a decimator afterwards. Or BlendBob plus a decimator, as I sometimes do.

And I don't speak for jagabo. He may have different opinions on some of the things discussed.

[jagabo]

Hopefully this isn't too elementary of a question but... So no matter what the original source on the ntsc LD is; Hollywood movie @24fps or a music video compilation that might be 29.97, the captured footage definitely could be improved by QTGMC? and if you're going through all the hassle of capturing it probably doesn't make sense to even use Yadif, it's QTGMC or nothing right? When using Vegas Yadif does seem to be a slight improvement over the built-in options of Blend/Interpolate fields.

As far as IVTC goes, does it matter if the original source is film or video? Just wondering because of the different frame rates.

And again, you're saying if deinterlaced by QTGMC, you may not need to do IVTC?

Aside from the Yadif vs. QTGMC comparison, I never said any of that. Most of it is exactly the opposite of what I said.

[clashradio]

Sorry jagabo. I got confused with when needing to deinterlace and/or IVTC between video or film sources.

[clashradio]

How can you tell if the footage on the LD is film or video? I have Iron Maiden Live After Death on both LD and dvd. The back of the dvd says it was shot on 35mm yet it looks like it was shot on video probably due to the U-matic or D1 transfer? One LD that I'm sure is video sourced is Thin Lizzy The Boys Are Back In Town. But I wouldn't be able to tell the Maiden show was on film if it wasn't stated.

[jagabo]

Find a section where there is a lot of motion. Step through the video field by field. If you see movement at every field the video is interlaced. With NTSC video, if you see a repeating pattern of 2 repeats, 3 repeats, it's telecined film. With PAL you may see motion only every other field. I usually use the Bob() filter in AviSynth. Note that that leads to some vertical bounce from field to field -- that doesn't count as motion. Yadif(mode=1) is good for this too.

[clashradio]

Ok, now I'm really confused. I captured (2) 5 second laserdisc clips. One from film source (Star Wars US 1992 Special Widescreen Edition) the other from video source (Thin Lizzy The Boys Are Back In Town Japan 1990 Live in Australia-might be PAL video tape?).

When viewing frame by frame in the Timeline, there is no pulldown pattern with either clip. I realize you may not be familiar with Vegas but I have the Properties set to: "field order"=LFF "deinterlace method"=none. I also tested with field order set to none.

But when I add Yadif as a Media Effect there is a repeated pattern. When I added Yadif as an effect I set Vegas' Properties field order and deinterlace method to none.

For Star Wars if I select Yadif's field order to LFF w/Parity also at LFF I get one repeating frame every 5. If I select UFF w/Parity UFF it repeats one frame every 4.

For Thin Lizzy there are no repeated frames when Yadif is added.

[jagabo]

Star Wars is telecined film. Thin Lizzy is interlaced video. Vegas is probably deinterlacing the timeline for your (people get freaked out when they see comb artifacts). Open your videos with VirtualDub2. Apply the Bob Doubler filter (or Deinterlace -> Yadif -> Double Frame Rate TFF). You'll see comb artifacts in the input pane, motion as described above in the output frame.

You may run into another problem with MJPEG. Some MJPEG encoders/decoder put the fields in the wrong position. You may need to specify the swap fields option in the decoder, or using a filter (VirtualDub has the Fields Swap filter, AviSynth SwapFields()). Look at diagonal edges in non-moving portions of the frame. If they are smooth you are fine. If they are jagged (comb like artifacts) you have the wrong field positions.

A still frame with fields in wrong position on the left, correct position on the right:



If you need to use both filters apply field swap first, bob doubler second. Note that field swapping also reverses the field order (order in which the fields should be displayed when bobbed) so you you'll may need to switch from TFF to BFF, or vice versa.

[clashradio]

I forgot to mention; MediaInfo reports both clips as UFF where Vegas states them LFF. Does that matter?

For Star Wars, when you say it's telecined film does that now become interlaced video when put on video tape/laserdisc/dvd?

Also I can see the interlace jaggies in the Timeline before I add Yadif .

[jagabo]

The different field orders probably reflect the fact that some MJPEG decoders swap field positions by default, others don't. Some of the early capture cards stored the upper field below the lower field. So instead of scan lines appearing in order within the frame (0, 1, 2, 3, 4, 5...) each pair of lines was swapped (1, 0, 3, 2, 5, 4...). Restoring the proper position also has the effect of reversing the field order (which is seen first when bobbed).

Analog NTSC video is always interlaced 59.94 fields per second. Anything that's not already that format must be converted to that format for broadcast or transmission (composite, s-video, etc.).

Note that with DVD, since it's digital frames can be encoded at 23.976 fps progressive. When that's done the video includes pulldown flags that tell the player how to produce 59.94 fields per second for analog output. Pulldown flags on DVD can be used for any progressive frame rate from 19.98 fps to 29.97 fps.

[clashradio]

Does it matter what the capture devices' field order is set to when capturing? All other capture devices I have never had this option. I recently bought a Motu HD Express pci and have been using the default setting of LFF.

[jagabo]

It really shouldn't matter whether you capture TFF or BFF. Since interlaced video is simply an alternating sequence of top and bottom fields you can start with a top field then add the next field to complete the frame (TFF), or start with a bottom field and add the next (BFF).

I would also use a lossless codec rather than MJPEG.

[clashradio]

I did a few quick tests; looks like it doesn't matter what field order I select when capturing, even Progressive.

The only two codec options I have are V210 & UYVY. I thought V210 was uncompressed? MediaInfo reports 4:2:2.

I just did two, one-minute tests. One with V210 & the other with UYVY. Both files are the same size of 1.55gb? MediaInfo does report the file captured with V210 as lossless. But it also reports the file captured with UYVY with a codec ID of v210.

[jagabo]

V210 should be 10 bit 4:2:2. UYVY is normally 8 bit 4:2:2. But the latter should be smaller. 1.55 MB/min is about the right size for uncompressed 10 bit UYVY and V210.

http://fourcc.org/pixel-format/yuv-uyvy/ (8 bit)
http://fourcc.org/pixel-format/yuv-v210/
https://web.archive.org/web/20120201095328/http://developer.apple.com/quicktime/iceflo...h019.html#v210

[clashradio]

I knew V210 was 10-bit, but didn't realize it is MJPEG. Do you have any suggests since Adobe only has the two codec options for capture?

I use Adobe for capture but Vegas for editing. Seems in order for the Motu to capture it needs Adobe, or at least for a PC, not sure about a MAC.