Elecard
October 23, 2014, 11:05:26 am *
Welcome, Guest. Please login or register.
Did you miss your activation email?

Login with username, password and session length
 
   Home   Help Search Login Register  
Pages: [1] 2
  Print  
Author Topic: Bug in Buffer Analyser for .264 streams  (Read 27580 times)
jreichel
Newbie
*
Posts: 2


View Profile
« on: August 24, 2006, 03:48:24 am »

Hi,
I found the followings bugs in the Buffer Analyser (which is a great tool otherwise) when used with .264 files:

The analyser uses  the cpb_removal_delay_length_minus1 instead of the initial_cpb_removal_delay_length_minus1 as specified in the standard to decode the initial_cpb_removal_delay.

The analyser ignores initial_cpb_removal_delay_offset. This can create buffer underflow when the delay is >0.

When multiple Buffering period SEI mesages are present in the bitstream, the anylyser behave corectly for the AU with the BP SEI, but the next NAL has a removal time which is not corect. (behave like the previous BP SEI was the first BP SEI, and restart all the counter for the removal time, which creates a buffer underflow)

Only the VCL buffer is analysed, how do you analyse the NAL ones if they are not the same ?

The zoom fonction sometime behave very strangly (zomming out give you a timeline on x from 0 to 0)

Julien

PS: I have some bitstreams wich can generate all those bugs if needed
Logged
vootshee
Newbie
*
Posts: 7


View Profile
« Reply #1 on: August 28, 2006, 03:48:51 am »

HI, upload your stream on our FTP

http://ftp://uploader:ybrjveybrf,tkmyjcnm@ftp.elecard.net.ru/
User: uploader,
Password: ybrjveybrf,tkmyjcnm.

This password gives you all necessary privileges.

Please, do not exceed file size limit (50 MB).
Logged
jreichel
Newbie
*
Posts: 2


View Profile
« Reply #2 on: September 04, 2006, 06:07:53 am »

The files have been uploaded in tzhe 4vootshee directory
BP_SEI.264 has two BP SEI
cpb_removal_delay_length_minus1.264 has different length for the CBP
initial_cpb_removal_delay_offset.264 has an offset !=0

Note that with the latest version of the StreamEye tools (2,1.0 bu 60706) it is not possible to study those stream has the analyser get some errors. Note that teh JM decodes such stream without any problems
Logged
Joerg Vogler
Guest
« Reply #3 on: January 08, 2007, 02:35:23 am »

Hi there,

I've investigated verious versions of the Buffer analyser, searching for a tool that can properly display the CPB usage of H.264 streams.

Version 1.0.16 (Build 60707, contained in StreamEye Tools 2.9), which I've been using for a while, displays the CPB usage of the streams but apparently contains the bug discussed earlier in this thread.

Therefore I tried the latest available version 1.0.31 (Build 61206, contained in StreamEye Tools 2.9.1). This version does not display the CPB usage of the identical streams at all.

Any comment?

JV
Logged
Joerg Vogler
Guest
« Reply #4 on: January 08, 2007, 02:45:49 am »

Hi again,

There seems to be another bug in Buffer Analyser 1.0.16 besides the ones discussed before:

According to subclause 7.4.1.2.3 of the H.264 specification, filler data NAL units belong the the access unit of the previous coded picture, e.g.:

Access unit 0: SPS, PPS, SEI, slice data, filler data
Access unit 1: SEI, slice data, filler data
Access unit 2: SEI, slice data, filler data
...

But the Buffer analyser seems to calculate the size of the filler data into the following access unit.

In streams operating at the upper limit of the CPB, this behaviour leads to an incorrect CPB overflow indication.

Unfortunately I could not confirm if this problem still exists in the latest version 1.0.31 due to the problem reported in my previous posting.
Logged
Nick
Elecard Team
*
Posts: 32


View Profile
« Reply #5 on: January 24, 2007, 06:38:29 pm »

Please, upload your streams on our FTP

http://ftp://uploader:ybrjveybrf,tkmyjcnm@ftp.elecard.net.ru/
Logged
Joerg Vogler
Guest
« Reply #6 on: January 25, 2007, 08:36:13 pm »

I've uploaded the file as requested. Its name is F1_QVGA.264. The stream contains 1 SPS, 1 PPS and 45 pictures (1 slice per pic), each preceded by an SEI unit with buffering and picture timing information for the HRD.

The pictures 2 to 12 are followed by a filler data NAL unit. If the size of the filler data is counted to the preceding pictures as specified in H.264, the CPB constraints are fulfillled.

But the Buffer Analyser 1.0.16 seems to count them with the following pictures and thus shows an overflow at pictures 3 to 13.

Further, as I've reported, the latest available Buffer Analyser version 1.0.31 (Build 61206, contained in StreamEye Tools 2.9.1) is not able to display the CPB usage at all.
Logged
Nick
Elecard Team
*
Posts: 32


View Profile
« Reply #7 on: January 29, 2007, 02:02:51 am »

Thank you for useful information. We'll try to scrutinize this issue as soon as possible
Logged
Nick
Elecard Team
*
Posts: 32


View Profile
« Reply #8 on: January 29, 2007, 03:32:38 am »

Try to change Elecard MPEG Demultiplexer (..\Elecard StreamEye Tools\Components\empgdmx.ax) in v1.0.31. You can use demultiplexer from old version of BAnalyzer (v 1.0.16) (..\Program Files\Common Files\Elecard\empgdmx.ax). Simply replace files.
Logged
Joerg Vogler
Guest
« Reply #9 on: February 26, 2007, 09:09:53 pm »

I just downloaded the latest build of the StreamEye tools (v2.9.1 build 70216), added the old filter from version 2.9 as instructed, and apparently the CPB analysis including filler data runs well  Smiley

But the handling of initial_cpb_removal_delay_length_minus1 and cpb_removal_delay_length_minus1 still seems to be faulty :?

If both values have the same value (both 17 in my test), the buffer analyser works well. But if they differ (initial_cpb_removal_delay_length_minus1 = 17, cpb_removal_delay_length_minus1 = 7), the buffer analyser shows some misbehaviour when I zoom in and out...
Logged
Nick
Elecard Team
*
Posts: 32


View Profile
« Reply #10 on: March 04, 2007, 09:38:20 pm »

it is possible, that you use old version of bAnalyzer.dll .  Try to register in system new dll.  ("regsvr32 bAnalyzer.dll") or reinstall new banalyzer, and after that replace filter (without reinstall old version of application).
Logged
Joerg Vogler
Guest
« Reply #11 on: March 04, 2007, 10:19:59 pm »

I've registered the DLL. But the error still occurs.

BTW, I had uninstalled the entire old version of the StreamEye tools and kept only the old filter empgdmx.ax. Then I installed the new version and copied the old filter into it.

But just now I see that there are still many files in Program Files\Common Files\Elecard... Probably they are from the Elecard MPEG player or AVC plug-in that I have installed as well.

I'm now cleaning up everything and installing the latest versions of all applications. Hope this will solve the problem.
Logged
Joerg Vogler
Guest
« Reply #12 on: March 04, 2007, 10:43:15 pm »

I've uninstalled the MPEG player, the AVC plug-in and the StreamEye tools, deleted any remainders in the directories Program Files/Elecard and Program files/Common Files/Elevard. Then I installed the latest versions of all these applications.

Now the buffer analyser works well, even without copying the old demultiplexer filter into it, as long as initial_cpb_removal_delay_length_minus1 and cpb_removal_delay_length_minus1 are equal.

But if these two HRD parameters differ from each other, the zoom-out function of the buffer analyser misbehaves again. I've tested it with a .264 stream of 10 seconds (150 frames at 15 fps).

When I open this stream in the buffer analyser, it shows 3.64 seconds of the stream correctly. The first zoom-out changes to 7.01 sec, which still seems to be correct. The second zoom-out should show the whole 10 seconds.

This works well with a stream having equal initial_cpb_removal_delay_length_minus1 and cpb_removal_delay_length_minus1. But ith they differ, the second zoom-out zooms in to the first 1.83 seconds of the stream.

I can offer you the streams for testing if it helps.
Logged
Nick
Elecard Team
*
Posts: 32


View Profile
« Reply #13 on: March 05, 2007, 10:33:51 pm »

Ok, upload your stream.

http://ftp://uploader:ybrjveybrf,tkmyjcnm@ftp.elecard.net.ru/
Logged
Joerg Vogler
Guest
« Reply #14 on: March 05, 2007, 11:49:09 pm »

I'm sorry, but I could not log in to your FTP server using the given user name and passord (well, I simply clicked the link and got that error message).

Can I send you the streams by mail? I have two streams of 470 kB each.

Joerg
Logged
Pages: [1] 2
  Print  
 
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.16 | SMF © 2011, Simple Machines Valid XHTML 1.0! Valid CSS!