At 8:58 AM -0700 9/21/05, dgrimes@xxxxxxxxxxxxxxxxx wrote: >"The decision to increase GOP size when using H.264 should be based on >image quality objectives"--Craig > >Thanks, Craig, for the detailed response. I definitely agree that GOP >should be based on the media for the best quality and coding efficiency. >It would be great if GOP didn't have to be fixed but rather could be >determined scene by scene based on the factors that require a new I-frame. >I know that some encoding software will force an I-frame at scene >transitions, but other things like dissolves and fades are not addressed >(at least not in software I own). And there probably isn't the time to >determine the best placement of I-frames in a live encoder (without a long >delay). Actually it is no that difficult for a real-time encoder to force I-frames, IF... The major issue is analysis of the source video so that the encoder has foreknowledge of what it about to do. Many MPEG-2 encoders now use multi-pass techniques to do exactly this. One chip is used to gather statistics which are then used to guide the second encoder chip that does the actual encoding. The biggest drawback with this approach is that it introduces more latency in the encoding process. An I-Frame can be forced whenever the first encoder sees a spike due to a scene change of a major change, like an explosion. Someday, encoding will become an integrated part of the post-production process. We will store metadata to help guide the encoder. Scene changes will be a piece of cake. But for real time encoding the answer is to add enough latency to the encoding process to gather statistics and use them to do the best job possible. Regards Craig ---------------------------------------------------------------------- You can UNSUBSCRIBE from the OpenDTV list in two ways: - Using the UNSUBSCRIBE command in your user configuration settings at FreeLists.org - By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word unsubscribe in the subject line.