[Live-devel] H.264 in MPEG-2 TS streams roughly at the start

[Warren Young]

On 9/20/2012 5:11 AM, Warren Young wrote:
> libx264 is pretty much incapable of true CBR.  You have to do
> null-stuffing to get a CBR-like result, and even then, you end up with a
> fair bit of variability.  I have bitrate graphs to prove it. :)

Someone emailed me privately asking about those graphs.

They were created with an R script I wrote, since I wanted deep 
visibility into the graph creation process.  (I also wanted a second 
opinion on what I was getting from other bitrate viewer tools.)  I asked 
for and received permission to release that script publicly as free 
software.  It lives here now:

	https://code.google.com/p/etr-bv/

It's not as fast or as slick or as easy to use as some tools I've used, 
but when you run the script inside RStudio, you have an amazing amount 
of power to manipulate the raw statistics, customize the pretty 
graphics, or answer questions nongraphically.

etr-bv even has at least one feature I haven't seen in other tools: it 
gives min, max, mean *and* standard deviation for the overall data rate. 
  SD is useful in this CBR investigation because it helps you compare 
two encodes, to see which is "more constant."