IVTC is useless.

[sugarkang]

i'm doing a lot of encodes with anime, tv series, films... whatever.
so aside from doing a force film (yes i know this is a form of IVTC), the other kinds seem to be somewhat useless. i'll get to the point...

[1] i ripped the first 6 episodes of ROBOTECH. it was a "hybrid" disc. i thought "uh oh... nightmare"
usually i'd use this in my script:
telecide(post=true,blend=flase,guide=1,chroma=true)
decimate(mode=2)

but then i thought... why not just leave everything at 29.97, and just fast deinterlace, simple resize this bad boy. well, i did... and the results were spectacular. i get 3 episodes per 700MB disc, and the compressiblity is about 70% or higher.

but what about the frames i could be saving right? converting from 29.97 --> 23.99 would save lots of space right? not exactly. here's what happened this morning:

[2] i started work on the TV series "24" with kiefer sutherland. most episodes are between 90-95% film. i figured i'd just IVTC this sucker, until i found certain parts (split screens) which were interlaced. now here's the weird part:

i ran compressibility tests at 29.97 frames. fast deinterlace and simple resize. i get 78%+ at 608x376 res.

i ran compressibility tests for 23.97 frames [IVTC]. fast deinterlace and simple resize. i get 70% at 608x376 res.

is that weird or what? i mean of course it doesn't make sense. i should be able to have lots of extra space by discarding 6 frames, but it isn't so. hence, my new way of doing things has simplified a TON!! i'm going to forget about all these manual script edits, as they seem useless to me.

from now on if it's a true film source, i'll just force film.
but every other DVD, i'm just going to deinterlace, and choose simple resize, or sharp bicubic [depending on the space limitation].

does this make sense? or can someone explain what happened on my episodes of "24"?

[agent2099]

Interesting.

[jggimi]

I don't think you were testing compressibilty. Here's why:

Changing the framerate at the bottom of the Gknot window will not change the video stream. The .d2v file is the source for that, and changing that value will not change anything in the .avs script used for encoding.

What does changing that value do? It changes the number of frames in the b/p*f calculations, including % Load.

It uses the elapsed time of the video to calculate a new number of frames.

Example: If you reset any of the 8 "save" slots in Gknot, along with selecting DivX 3, you get the following:

1) The FrameRate is set to 25fps
2) Video length is set to 1 hour, 30 minutes (5400 seconds)
3) Resolution is set to 640x352.
4) Video size will be set to 713567 KB

Note the number of frames is 135000, for 25*5400 seconds

Adjust the Framerate to 30fps:

1) Bitrate drops from 1057 to 1056kbps (a rounding error only)
2) Length of video stays at 1:30 (5400 seconds)
3) Video size drops to 712921 KB (rounding error)
4) Frames jump to 162000, for 30*5400 seconds

Note that the Bits/Pixel*Frame has altered from 0.192 to 0.160, due to the number of frames changing from 135000 to 162000 in the b/p*f calculation.

[jggimi]

As a followup, here's the test I recommend you try -- but only with Telecined content that can't be force filmed. Something you would use Telecide/Decimate to reduce to 23.976fps.

For both tests, do a single pass 100% quality based encoding. Pick a short sequence, it should only take a few minutes:

First test: De-interlacing only, leaving at 29.97.
Second test: Telecide/Decimate, reducing to 23.976.

Compare the output file sizes....

I just did this with 1000 frames of content at 608x352, and just deinterlacing produced a file that was 19,190KB. IVTC produced a file that was 16,078KB, a reduction of more than 19%.

[sugarkang]

Changing the framerate at the bottom of the Gknot window will not change the video stream. The .d2v file is the source for that, and changing that value will not change anything in the .avs script used for encoding.

here's what i did exactly:

1. i loaded the .d2v in gknot. [original .d2v file was kept in 29.97 and not force filmed]
2. set res to 608x376 / auto cropped / smartcropped all
3. pressed save & encode button.
4. chose fast deinterlace / simple resize
5. ran compressibility test at 5% [changing to 100% will not do anything because it takes frames throughout the film. standard deviation laws make the calculation error less than 1%]

once test was done, the percentage should be correct. it showed up at 78.5%

then i closed out of gknot.

then redid steps 1 through 3.

4. chose IVTC / simple resize
5. ran compressibility test at 5%

once the test was done, the percentage changed to 70%. that percentage should be accurate. it's not from changing all these other parameters like using different slot projects like you're suggesting.

now i do recall that when i ripped ninja scroll, where 2 of every 5 frames are bad, that IVTC did give me similar results like you're saying. i think it did show up with a better percentage, but i'm not sure. i will test as soon as my other episodes of "24" are done. [i have them all queued up right now] and with great results, 232MB per episode, although it's pretty dark.

anyway, could you try this experiment on an almost pure film DVD? because you're using different source than i'm using. yours is telecined, whereas mine is a hybrid, although 90% of it is film.

[manono]

Hi sugarkang-

Always asking the tough questions, eh? I'll assume these are DVD sources-I don't know anything about capturing. When you run the compress test on a source that you IVTC, the results will show for the original .d2v (at 29.97fps in this case). To make the compress accurate for your IVTC'd material, you have to change the fps in the little box in the lower left corner to 23.976. It will automatically boost the percentage figure by 25%. So, if you got 70% based on the original 29.97fps .d2v, it should go up to 87.5% or thereabouts after IVTC. And that's the accurate figure.

I'm not positive why you got 78% with one and 70% with the other, but I suspect it has to do with the use of Fast Deinterlace in the former and not the latter. In addition, by keeping Film material at 29.97, it will play slightly jerky as there are duplicate fields still in there. This is best seen in panning shots. As for the hybrid material (mix of Film and Video), it's a tough decision as to whether or not to try and IVTC or keep it at 29.97fps. It depends on how the video material looks after IVTC and how much of it there is. But if you decide to keep it at 29.97fps, Decomb's Mode=1 is by far the best way to handle it as it preserves most of the Progressive frames and it will play "clearer" and smoother.

[sugarkang]

helpful as always manono.
maybe you should go *out* to the movies tonight! ;)

and btw. yes i changed the fps to 23.97, and got 70%.

and you're right about the field deinterlace... i'm getting much higher compression with that on, than without any field operations turned on. so i think that may be the culprit, although IVTC should run a default field deinterlace with telecide.

regarding leaving the duplicate frames on 29.97 film. i don't understand why it would play jerky. i mean, on a dvd since it's telecined to be 29.97 in the first place from film... wouldn't the source DVD play jerky also? when you're playing divx, it's just playing the original 29.97, just like a DVD i thought.

i haven't noticed any jerkiness in my rip of "24" which was 90% film, but i can't be sure... maybe i missed it.

[manono]

Hi-

So you did switch back to 23.976fps. Then FastDeinterlace (aka VerticalReduceBy2) plays a bigger part than I thought. Since it cuts the vertical resolution in half (later corrected by resizing), half of the video information is lost, so perhaps that explains it (the resulting video should be noticeably inferior also). This is from Doom9's DVD2SVCD Guide (http://www.doom9.org/mpg/dvd2svcd.htm). "If you need a quick and dirty deinterlacing select SeparateFields/SelectEvery or VerticalReduceBy2. Telecide is a more thorough operation and will give you the best results." So, I think you should run your tests with the same deinterlacer. That is to say, when doing your 29.97fps test, use FieldDeinterlace because, as you said, FieldDeinterlace is default in Telecide(). Then it should be more of an apples to apples comparison, and the resulting percentages should be more in line with what we expect.

jggimi's test accomplished pretty much the same thing (he didn't say which deinterlacer he used, but I suspect it was the FieldDeinterlace for both tests).

Regarding jerkiness-while leaving Film material at 29.97fps, as opposed to Force Filming or IVTCing it, won't show jerkiness as obviously as Force Filming or IVTCing video material, it's there (like I said, the best way to see it is to find a panning shot). Or you can open the resulting .avi in VDub-Nandub and scroll to a motion scene and advance frame by frame, and you'll notice the slight pauses from time to time. You're right-it's not very noticeable when watching the DVD on a standard CRT television. I think it's partly because we're used to it, and partly because it's very slight (and maybe because TVs blur everything anyway). It'll be less noticeable on a TV playing fields than an .avi playing frames. But I refer you back to the picture in the tutorial where they're repeating 2 fields in every 10. You can see the same thing yourself by adding a SeparateFields() command to your .avs (using no Field Operation in DVD2AVI and no IVTC in the .avs) and then opening it in VDub. There are repeated, duplicated fields. It has to play slightly jerky. That's what a TV does-plays it as if there's a SeparateFields() command in there (at 59.94 fields per second). The best way to tell the difference is to play the same DVD through a Progressive Scan DVD player and onto an HDTV or through a projector (set to 72HZ perhaps, but not too many people have that kind of setup yet).

when you're playing divx, it's just playing the original 29.97, just like a DVD i thought.

If the source was originally Film, then it was originally 24fps. The telecine process adds all those extra fields so it can play on NTSC televisions. Film is actually stored on a DVD at 24fps. The telecining is done with Flags or instructions which tell the DVD player to send 59.94 fields per second to the TV set. The extra fields are added by the DVD player when sent to the TV. Standard NTSC TVs can only output 59.94 fields per second. And Progressive Scan DVD players and HDTV will soon change all that. Force Film or IVTC returns it to the way it's stored on the DVD. Robshot's Article (http://www.doom9.org/synch.htm) on this site explains more about the process. Damn-that was a long one. Quit making me think so much.:)

[sugarkang]

thank you very much manono. i hope someday i can stop learning about gknot, and get on with my life! ;)