carmiyu
15th December 2005, 08:09
Dear experts,
When i read H.264 spec, i feel that the POC mode 0 has bad error resillience capability. If one or more reference frames got lost, the rest picture order counts might all be wrong. Is POC mode 0 proposed for error-free environment only??
I searched in Doom9 forum and found out an old thread talking about this. In the thread, a JVT document was posted :
"http://ftp3.itu.ch/av-arch/jvt-site/2002_12_Awaji/JVT-F044.doc"
In this document, a paragraph is as below:
*********************************************
'We propose a POC coding mode that is very close to the POC coding mode 0 of the Klagenfurt output version of the standard (JVT-D157, JVT-E022d7). The POC coding mode 0 of JVT-D157 was removed obviously as a response to the USNB comment in JVT-E68r1:
"In mode 0, the decoder needs to maintain the correct value of PicOrderCntOffset in order correctly decode picture order count. If more than two or more picture are not decoded (through error loss or through the ability to skip frames provided by enhanced GOPs) then it possible for the value of PicOrderCntOffset to become incorrect. Unlike POC mode 1, this can happen even if the pictures are not stored. This breaks the property of the AVC decoder design that the decoder does not need to decode non-stored frames to correctly maintain its state."
We believe that the POC coding mode 0 of JVT-D157 was misunderstood in this comment. In this paper, we propose the POC coding mode 0 again and show that it actually is very error-resilient. We have made minor modifications on the terminology, aligned the proposed text with the latest draft standard, and removed the dependency on non-reference pictures.'
**********************************************
In the paragraph above, the author mentioned that POC coding mode 0 is very error resilient after some minor modification. But in my humble opinion, the minor modification doesn't change anything at all. I still believe in POC coding mode 0, if one or more reference frames are lost, the picture order counts of the rest frames are very likely all be wrong. Am i correct on this?? Please kindly share your comments. Thanks!!
SL
When i read H.264 spec, i feel that the POC mode 0 has bad error resillience capability. If one or more reference frames got lost, the rest picture order counts might all be wrong. Is POC mode 0 proposed for error-free environment only??
I searched in Doom9 forum and found out an old thread talking about this. In the thread, a JVT document was posted :
"http://ftp3.itu.ch/av-arch/jvt-site/2002_12_Awaji/JVT-F044.doc"
In this document, a paragraph is as below:
*********************************************
'We propose a POC coding mode that is very close to the POC coding mode 0 of the Klagenfurt output version of the standard (JVT-D157, JVT-E022d7). The POC coding mode 0 of JVT-D157 was removed obviously as a response to the USNB comment in JVT-E68r1:
"In mode 0, the decoder needs to maintain the correct value of PicOrderCntOffset in order correctly decode picture order count. If more than two or more picture are not decoded (through error loss or through the ability to skip frames provided by enhanced GOPs) then it possible for the value of PicOrderCntOffset to become incorrect. Unlike POC mode 1, this can happen even if the pictures are not stored. This breaks the property of the AVC decoder design that the decoder does not need to decode non-stored frames to correctly maintain its state."
We believe that the POC coding mode 0 of JVT-D157 was misunderstood in this comment. In this paper, we propose the POC coding mode 0 again and show that it actually is very error-resilient. We have made minor modifications on the terminology, aligned the proposed text with the latest draft standard, and removed the dependency on non-reference pictures.'
**********************************************
In the paragraph above, the author mentioned that POC coding mode 0 is very error resilient after some minor modification. But in my humble opinion, the minor modification doesn't change anything at all. I still believe in POC coding mode 0, if one or more reference frames are lost, the picture order counts of the rest frames are very likely all be wrong. Am i correct on this?? Please kindly share your comments. Thanks!!
SL