DivX DRF Analyser: The deviation should be less than 1?

[Randm]

what does the standard deviation mean? why should it be less than 1...i've searched all over & can't find any explanation or reason why...

[manono]

It seems to be a function of how much the quants vary from the average. I opened an AVI I had done using a CQ of 3 with 1 B-Frame and the Standard Deviation was 1 (the I and P-Frames were all quant 3 and the B-Frames all quant 5). I redid it, this time using a CQ of 3 but with B-Frames disabled. The resulting SD was 0 (all frames were quant 3). Then I did a couple of 2 passes, one of high quality with quants allowed to vary between 2 and 31 (roughly the same size as the source) and the SD was 1.02 with the quants varying between 2 and 5. Then I did another 2nd pass with a much smaller size and the SD was 1.47, with the quants varying between 2 and 8.

It's looking for high quality encodes, without much quant deviation from the norm. One way is to do 1-pass encodes, using a reasonably decent low quant. Another is to do 2-pass encodes and severely limit the quant range. For example, when doing 2-pass encodes I usually limit them to between 2 and 4. AutoGK uses 2-6, usually. Using 1-31, or even 2-31 is a pretty bad idea anyway, I think. Doing it this way requires a fair knowledge of the final size in relation to the quant 2 first pass size, which you can learn by saving a playable first pass.

Whether or not you agree with keeping the SD around or below 1 is up to you, but I do think that having too many high quant frames in the encode is bad for the overall quality. It's just one metric to take into account when trying to figure if the encode is any good or not. Finally, I did a 1-pass quant 10 encode with B-Frames disabled. The SD was 0 and, of course, the encode looked horrible.