B-frames... Benefit of 1.62 ratio and 0.00 offset?

[Ranguvar]

In the past, I've seen the b-frame quantizer ratio of 1.62 and offset of 0.00 recommended over the default of 1.50-1.00. Can anyone tell me what the benefits are; how will this affect the video?

Making an educated guess, it's so that b-frame quantizers are even closer to the reference frame's quantizer at low quants, and farther away at high quants... don't fully understand why that would be desirable, though.

And I've heard ratio of 1 and offset of 1 recommended for anime... any thoughts?

[henryho_hk]

1) 1.62 is nearest to the Golden Ratio (~= 1.6180339887...) [The average ratio is lower than that bcoz, from my limited memory, the final step is truncation but not rounding.] [sysKin said (http://forum.doom9.org/showthread.php?t=115357) that it's indeed rounding.]
2) (1.62, 0) gives smaller B-quant unless the I-quant get bigger than 8 (which means a terrible picture itself - we're always better off using a smoother matrix than higher quantizers). Hence, the B-frames are better looking then (1.5,1) for most of the times.
3) For animes, in addition to (1,1), it is also recommended to use the smooth(er) H.263 other than matrices like MPEG.

[Sagittaire]

Well only one word ... ridiculous!

IMO XviD by default use overquant for bframe. 1.00/1.00 for ratio/offset produce better quality for my eyes.

[Brother John]

[sysKin said (http://forum.doom9.org/showthread.php?t=115357) that it's indeed rounding.]
He should have said quants are always rounded down to the nearest integer. At least that’s the behaviour I got from practical tests. I prefer 1.62/0 because it’s less aggressive than Xvid’s already aggressive B-frames. For high bitrate/low quant scenarios 1/1 is more or less the same. And even 1/0 can work well.

Here’s a comparison between 1.5/1 (»Standard«) and 1.62/0 (»Didée«). »P-Quant« on x-axis means average quant of reference frames.

http://img165.imageshack.us/img165/1795/xvidbframesdideemq6.png