NAL HRD has been committed!

[shon3i]

Originally Posted by mp3dom

I'll investigate tonight, but I had a buffer underflow in Scenarist today in a clip of 3 mins (it stops after 6 seconds of mux) with an avg bitrate of 25 Mbps (which is quite 'standard') and only one audio (DD at 640Kbps). It could be the vbv-bufsize set to 30.000 so tonight I'll try with a more conservative value.

What exact settings you use for buffer and max rate? I discovered in past (on earlier HRD patches aslo) that if use buffer different that max rate, scenarist discover potential buffer uder/overflow and stop muxing. I found if i set VBV 30000/30000 or 15000/15000 or anything same for buffer and maxrate scenarist mux flawlessy. IIRC CBP removal delay should be always 0.9, i don't know is that same as x264 VBV-Int option, which i try to adjust, to achieve that delay, but without success, i even get VBV errors durring encode . I read somewhere in scenarist documentation that delay should be always near 0.9. Other encoders (Cinevision, MC, Elecard) keep that on 0.7-1.0 nevemind i set for buffer/maxrate, while with x264 can be anything, for example if i set 30000/40000 it will be around 0.5-0.6, if buffer is higer will be over 1.0 and around 1.80

[mp3dom]

I have used 30.000 for vbv-bufsize and 40.000 for vbv-maxrate which, if I'm not mistaken, are also the standard values. I haven't tried the CBP delay, nice suggestion, I'm trying it!

[kieranrk]

Can you upload the encoded clip somewhere (or send it to me privately) so I can see if it's a real buffer underflow or Scenarist being wrong.

[mp3dom]

You have a PM.

[shon3i]

Originally Posted by mp3dom

I have used 30.000 for vbv-bufsize and 40.000 for vbv-maxrate which, if I'm not mistaken, are also the standard values.

Exactly, but removal delay will be 0.6 in that case which is maybe not recommended.

Originally Posted by kieranrk

Can you upload the encoded clip somewhere (or send it to me privately) so I can see if it's a real buffer underflow or Scenarist being wrong.

You not find any underflows in stream itself, its problem in VBV delay that must be some constant value. I will post some examples, of different encoders

[mp3dom]

Anyway it's kinda strange that I've encoded full Elephants Dream with even higher settings (same vbv_bufsize/vbv_maxrate but avg bitrate raised from 25Mbps to 30Mbps) without any problem (and it's a 10min clip where the clip that fail to mux is only 3 minutes and scenarist fail after 2 seconds!)

[shon3i]

Originally Posted by mp3dom

Anyway it's kinda strange that I've encoded full Elephants Dream with even higher settings (same vbv_bufsize/vbv_maxrate but avg bitrate raised from 25Mbps to 30Mbps) without any problem (and it's a 10min clip where the clip that fail to mux is only 3 minutes and scenarist fail after 2 seconds!)

What exactly settings? you use?

[mp3dom]

keyint at 24, min keyint at 2, bpyramid, 4 reference, 3 bframes, 4 slices, 30 mbps of bitrate, aq 1, all partitions, weight prediction at 2 and so trellis

[shon3i]

I mean VBV

[mp3dom]

Ah, sorry ^^; I'm using always 40K and 30K for maxrate and buffersize respectively.

[shon3i]

Originally Posted by mp3dom

Ah, sorry ^^; I'm using always 40K and 30K for maxrate and buffersize respectively.

Can you try with 30K for both max rate and buffer, you will not loose any quality anyway. Just for test. Because CBP removal will be then 0.9

[mp3dom]

Tested with both vbv values at 30K and Scenarist always throws a buffer underflow...

[kieranrk]

Well the file you sent me is valid. The H.264 spec only specifies a cap for the initial_cbp_removal_delay in VBR mode.

[mp3dom]

So is Scenarist wrong or there's really a underflow? It complains at about 2 secs.

[kieranrk]

Originally Posted by mp3dom

So is Scenarist wrong or there's really a underflow? It complains at about 2 secs.

There's definitely no underflow.

[mp3dom]

From a certain point of view I'm glad, from the other I'm a bit sad because I can't find a way to let Scenarist accept that footage encoded with x264. Anyway, thank you very much for the patience

[shon3i]

mp3dom can you send me source and encoded clip, for testing on PM?

[mp3dom]

The original clip is too huge for my adsl line (we are speaking of 2.7GiB of a YV12 video already compressed in lossless way, also is a working project so I can not redistribute it). I can send the portion that I've already sent to kieranrk and it's about 5 sec of video but I doubt that it's possible to reproduce the underflow (or the problem) with few seconds...

[shon3i]

Ok, can send then 5 sec of source and encoded video which fail to mux?

[shon3i]

Btw i found rule about VBV Buffer in Cinevision manual.

Bitrate buffer specifies the size of the virtual buffer verifier in bytes or bits. This value
should be adjusted to bitrate (Constant Bitrate) or to maximum bitrate (Variable Bitrate), to
avoid DTS/PTS underflows during muxing.

So if stream itself not contain any underflow, scenarist can detect possible underflow that not exist, but can result incorrect playback on SAP.