MPEG2 Encode - Scene Change Detection (Apple Compressor vs Adobe vs other)

[rallymax]

I'm trying to encode a DVCPROHD short film for release on DVD and Bluray.

It was authored in Final Cut Pro and for simplicity I had the film rendered on that MAC to MPEG2 (m2v) using Apple Compressor with the settings 8.5Mbps VBR 2-pass. The result was a .m2v file with an average of 7.2Mbps which would suggest that it was able to encode with 100% quality in less that the desired bitrate.

BUT.... The GOP structures don't detect the scene changes. As a consequence I can't put dvd chapter markers on the start of scenes. So I either have a flash of 1..5 frames from the prior scene or miss the first couple of frames into the scene when jumping from the "Scene Selection" menu to that chapter.

Is there another tool to encode with (eg AME (ie Main Concept) or FFmpeg or other?) that will do a better job at putting the full GOP frames on those boundaries?

thx.

[aedipuss]

gops fall where gops may. you can't change them on the fly in the middle of a video, they aren't variable. you can try moving the chapter points in your dvd authoring program as that's where they will be set for the dvd anyway.

[poisondeathray]

Chapter points require an I-frame. You would normally give your encoder a chapter list, so it forces an I-frame at that exact point

If your dvd authoring program is doing the encoding as well, then there should be no problem either (chapter points should correlate with the encode)

[manono]

Chapter points require an I-frame. You would normally give your encoder a chapter list, so it forces an I-frame at that exact point

In my opinion that's not such a good idea. It'll force a chapter point at that exact place, yes, but if the DVD is progressive it can also force it to drop to video sometimes if the frame you designate as a chapter stop isn't at the beginning of the 5 frame 3-progressive-and-two-interlaced pulldown cycle. That is, you can get a couple of frames of interlacing at that place. I don't believe that is normal practice for the big boys. Except for Criterion which is known for that problem. And because many, if not most, DVD players need a few frames to react, that interlacing may then be seen on a progressive display. Here's a review of a Criterion release that focuses on the problem. Scroll down to the 4 pictures in a row about a third of the way down the page:

http://www.mastersofcinema.org/bresson/Words/Criterion_AuHasardBalthazar.html

I use CCE and set the chapter points at the authoring stage and CCE seems to have good scene change detection so there aren't usually any problems setting the chapter points.

The result was a .m2v file with an average of 7.2Mbps which would suggest that it was able to encode with 100% quality in less that the desired bitrate.

If you want a larger size (and better quality) use a better quantisation matrix. I have no idea whether or not the Apple Encoder has that ability.

[poisondeathray]

manono - I've heard that argument before, and you have way more experience than me

Theoretically , let say CCE (or any encoder) detects a scene change and it isn't at the beginning of the pulldown cycle - what does it decide to do ?

[manono]

No problem. Ones encoded as progressive with the chapter stops set during authoring can (but not always do) show as 100% film in DGIndex. The ones with the chapter stops set during the encoding won't be 100% film. I'm ignoring the opening logos and end credits which can be partially or entirely video. For example, here's a part of a D2V (from Criterion's Seven Samurai), showing the drop to video at a chapter point:

900 5 0 662790144 0 1 6 71 72 d3 f0 f1 e2 f3 f0 e1 f2 f3 e0
900 5 0 663357440 0 1 6 71 72 d3 f0 f1 e2 f3 f0 e1 f2 f3 e0
900 5 0 663955456 0 1 6 71 72 d3 f0 f1 e2 f3 f0 e1 f2 f3 e0
900 5 0 664553472 0 1 6 71 72 d3 f0 f1 e2 f3 f0 e1 e2
900 5 0 665098240 0 1 7 92 b2 a2 b2 b2 e3 f0 f1 e2 f3 f0 e1 f2 f3 e0
900 5 0 665810944 0 1 7 71 72 d3 f0 f1 e2 f3 f0 e1 f2 f3 e0
900 5 0 666429440 0 1 7 71 72 d3 f0 f1 e2 f3 f0 e1 f2 f3 e0

Where you see those 2 2 2s, that's the chapter change and the drop to video and where the interlacing is that you'll see if there's movement at that point. If there were a problem with scene changes, then that sort of thing would be happening all the time, but it doesn't. It's only when you try and force chapter stops where there wouldn't ordinarily be an I-Frame that you get the problem.

This is also one good reason why it's not a good idea to use Forced Film in DGIndex unless it's 100% film. I made the D2V for a single VOB of the second disc of Seven Samurai and the film percentage at the bottom was 99.92. If planning to reencode the usual advice is that it's OK to use Forced Film if the film percentage is 95 or higher. But that's not true. This particular VOB (of 6) had 3 such places. Maybe it was during a black stretch between scenes or maybe it was at a scene with no movement, so you wouldn't notice any interlacing. I didn't check. But, unless it's 100% film it's usually a good idea to use 'Honor Pulldown Flags' and IVTC it. Because of the cadence break at such places, a frame or two is liable to be deinterlaced, but it's still better, in my opinion, than letting interlacing slip through.

[poisondeathray]

Thanks for explaining. Some of this is over my head. I booked marked some posts where neuron2 explained what the syntax means for DGIndex's d2v file... I'll have to go back sometime and digest it...

I understand the part about the problems with cadence breaks (and not 100% film) , but I'm failing to see why it would be different if the encoder detects a scene change and decides to place a keyframe (and it happens to be a chapter point when you author) vs. you force it to place a keyframe there at the same place. It's either something ridiculosly simple I'm missing or I'm going to have to dig up those posts and read the DGIndex manual

You say where "where there wouldn't ordinarily be an I-Frame that you get the problem." , is that the key ?

[manono]

OK, I see. rallymax's problem is that the I-Frames aren't set on scene changes for some reason. Therefore he can't set chapter points where he wants during authoring. Maybe Criterion has the same problem (poor scene change detection during encoding), but they're trying to set them during the encoding and wind up setting them on a scene change but also in the middle of a GOP. Or maybe the author of that review in that link above is correct in speculating that Criterion is using interlaced masters for encoding, and is both setting chapter points and IVTCing at the same time:

Also, forced I-Frame insertion at the encoding stage can significantly alter the MPEG structure, resulting in an IPBP sequence (instead of the nominal IBBP), where the (predictive) P-Frames actually may be referencing an interlaced frame in the original material, if the original material was interlaced. The "interlaced" feel of the compromised frames are, we think, a tell-tale sign that Criterion may have based their transfer on interlaced material.

I originally didn't think he was right, but your point makes me rethink it. I've gone through several Criterions chapter-by-chapter using PowerDVD with the deinterlacing set to 'Force Weave'. One can easily understand getting interlacing with chapter stops set during a dissolve. But I've also seen a ton of them set during real scene changes also with interlacing.