interlaced x264 encoding: warning

[WorBry]

I'm encoding anamorphic PAL (576/50i) DV to interlaced x264.mkv with MeGUI using the Standalone-Blu-Ray profile. Since the source is BFF I've added a (nal-hrd) --bff flag to the command line (with the 'interlaced' option unchecked) and changed the SAR to 64:45 for a 16:9 DAR. I've also tried reversing the field dominance of the source to TFF and encoding with the 'interlace' option checked (or unchecked, with nal-hrd --tff flag), with virtually identical results. However, in all cases I am seeing the following warning in the log file:

"Warning: Interlaced + Direct = Temporal is not implemented"

What is the significance of this warning?

Also, regarding the vbv-bufsize and vbv-maxrate. The Blu-ray profile has them both set at 40000, but in several posts I've come across about interlaced x264 encoding (with nal-hrd patch) at SD resolutions, folks are citing lower values. Any recommendations on this? If it's of significance, with my (denoised) DV material I'm generally encoding at around 7000 kbps to achieve a QP (I) of around 18.

[kemuri-_9]

However, in all cases I am seeing the following warning in the log file:

"Warning: Interlaced + Direct = Temporal is not implemented"

What is the significance of this warning?

it means exactly as it says:
you can't use --direct temporal or auto with interlaced material at this point in time (dunno if they plan to support it ever or not)...
in this situation, it switches over to spatial instead of what you told it.

[LoRd_MuldeR]

Anyway, this shouldn't be a problem. Using the "auto" mode I usually end up with something like this:
x264 [info]: direct mvs spatial:99.9% temporal:0.1%x264 [info]: direct mvs spatial:91.5% temporal:8.5%

(However I'm not sure whether interlaced video would benefit more from "temporal" than progressive material does)


Also, regarding the vbv-bufsize and vbv-maxrate. The Blu-ray profile has them both set at 40000, but in several posts I've come across about interlaced x264 encoding (with nal-hrd patch) at SD resolutions, folks are citing lower values.

If you are targeting for Blu-Ray, you should use the VBV settings suggested by the "Blu-Ray" profile. At least until you encounter any problems with these values.

Using lower VBV values will potentially hurt quality, which is unnecessary, as long as Blu-Ray is able to handle the higher values...

[Dark Shikari]

Anyway, this shouldn't be a problem. Using the "auto" mode I usually end up with something like this:
x264 [info]: direct mvs spatial:99.9% temporal:0.1%x264 [info]: direct mvs spatial:91.5% temporal:8.5%Of course, I have sources that get 100% temporal or close-to-it (parkrun, Touhou, etc) ;)

[nm]

If you are targeting for Blu-Ray, you should use the VBV settings suggested by the "Blu-Ray" profile. At least until you encounter any problems with these values.

Using lower VBV values will potentially hurt quality, which is unnecessary, as long as Blu-Ray is able to handle the higher values...
I've heard that Blu-ray limits SD video to High@L3.1. Some DXVA implementations may also require it. So, it would be safer to use the VBV settings from MeGUI's DXVA-SD-HQ profile.

[WorBry]

Thanks guys,

Anyway, this shouldn't be a problem. Using the "auto" mode I usually end up with something like this:
x264 [info]: direct mvs spatial:99.9% temporal:0.1%x264 [info]: direct mvs spatial:91.5% temporal:8.5%


Yes, I did some parallel encodes with my DV sources deinterlaced (TempGaussMC) to 50p and typical temporal values ranged from 0.1 to 4.0%; the higher values were more associated with frequent scene changes.

It's interesting to note that the 50i and 50p encodes acheived quite similar QP vales at the same bitrates, which I guess goes to show how relatively inefficient interlaced x264 encoding is. Do other H264/AVC encoders, like Nero (Ateme), and MainConcept, share this characteristic? I know that MBAFF in x264 is not truly macro-block adaptive (i.e. treats all macro-blocks as interlaced), but that should not make any difference for a pure interlaced source.

On another side note, it appears that the Yadif deinterlacer in FFDShow has some problem with interpretation of the field order; both bff and tff flagged x264 encodes stutter/flicker in double-rate (bobbed) mode when the field order is set to 'auto' detection i.e. it seems to apply the 'opposite' field order. The bob-deinterlacer in CoreAVC, on the other hand, now seems to get it right.

I've heard that Blu-ray limits SD video to High@L3.1. Some DXVA implementations may also require it. So, it would be safer to use the VBV settings from MeGUI's DXVA-SD-HQ profile.

Thanks nm. I'll leave the VBV settings as is, for now, but will bear this in mind if I encounter any compatibility/playback problems.

[Dark Shikari]

It's interesting to note that the 50i and 50p encodes acheived quite similar QP vales at the same bitrates, which I guess goes to show how relatively inefficient interlaced x264 encoding is. Do other H264/AVC encoders, like Nero (Ateme), and MainConcept, share this characteristic? I know that MBAFF in x264 is not truly macro-block adaptive (i.e. treats all macro-blocks as interlaced), but that should not make any difference for a pure interlaced source.Yes it should. Most macroblocks in an interlaced source, especially in low-motion scenes, are best coded as progressive.

In practice, the efficiency benefit from MBAFF is about 5-15%. From my own tests, it doesn't seem to be nearly enough to make Ateme's encoder better than x264, but it definitely helps.

[WorBry]

Yes it should. Most macroblocks in an interlaced source, especially in low-motion scenes, are best coded as progressive.

That's an eye-opener...for me at least.