Formula for encoding flvs

[stembridge]

I was given the following formula perhaps a couple years ago by someone...
video bitrate = 0.00022458 * width * height * framerate

Anyone else found this useful?

[tomlee59]

I've never seen that formula, but its functional dependencies seem not quite right -- it scales linearly with width*height, rather than, say, logarithmically with the product. But for some range of parameters, I'm sure that it can do ok, given that quality is a moderately weak function of bitrate beyond some minimum threshold.

[Case]

The value 0.00022458 seems to be what ffmpegX 0.0.9x r2 (*) uses for [Best]:

207 = 206.972928 = 0.00022458 * 320 * 240 * 12
703 = 703.0042472448 = 0.00022458 * 480 * 272 * 23.976
2328 = 2328.44544 = 0.00022458 * 720 * 576 * 25

(*) 0.0.9x has a working [Best] button for the ffmpeg engine, where 0.0.9y shows the iframeflag error (on my system).

[stembridge]

What about ffmpegx 0.0.9y?
What would be the difference? This has really been a useful resource; but I obviously didn't understand the underlying math!
Thanks!

[tomlee59]

I think what Case is saying is that version y should be using the same math, but it has a bug. No need to understand the underlying math more than this: higher framerates and larger resolutions should require higher bitrates. The approximate formulas used by ffmpegx assume that the bitrate is proportional to the simple product of these factors. The magic proportionality constant just fixes up the numbers to give you an absolute bitrate. I'm sure that it was chosen somewhat by eye, guided by some science.

DSP experts would point out that the assumption of simple proportionality is a bit dubious, but for a practical range of parameters, it yields a useful rough guide.

[Case]

What about ffmpegx 0.0.9y?

The same as before. The developer didn't change the suggested best bitrate for any 'old' presets. (Only the iPhone preset was new to 0.0.9y.)

... the underlying math

The 'best' bitrate is an optimum where a significant higher bitrate wouldn't improve so much and a significant lower bitrate would be almost unwatchable. 'Best' is more best choice than best quality.

So, one could get the multiplier by experimentation (testing various bitrates) and reversing the equation:
video bitrate = multiplier * width * height * framerate
multiplier = video bitrate / ( width * height * framerate )
One could keep it all mathematical and use "Peak signal-to-noise ratio" (psnr) as quality measurement, even across codecs, and avoid subjective human quality perception, although the threshold would still be subjective.

But if you think 207 kbps at 320x240 @ 12 fps still looks crap, then you're free to use a higher bitrate. The 'standard' used for FLV bitrates is pretty low, after all...