这两种2 pass编码模式哪个更好？
2 pass编码可以是（普通1 pass）→2 pass，也可以是（1 pass crf）→2 pass，这两种压法哪种质量更好？ 1 pass均使用了slowfirstpass
-
Billy Herrington
- 帖子： 83
- 注册时间： 2013-01-09 9:54
Re: 这两种2 pass编码模式哪个更好？
There's no difference, unless your 1st pass bitrate varies too much and thus the logged qp in stats file differs, which may affect the qp-bitrate-cost estimation in the 2nd pass.
Because abr and crf use exact the same rc method, except for that abr adds a qp penalty system based on encoded bitrate cost. And 1st pass stats file logs only a rough qp-cost estimation near the encoded 1st pass' bitrate, so for example the stats file of a qp=19 slice may be identical to the same one of qp=20 in the 1st pass. So only if the bitrate or qp-penalty gives a too large gap will those two different rc modes in 1st pass have real effects.
multi-pass generally logs qp-bitrate curve in the early pass, and estimates the real qp-bitrate curve for the following pass, according to which the qp-offset is calculated to match the target bitrate. And if the estimation differs from real encoded one, a cost penalty will be added to the following frames to be encoded. So the problem in multi-pass encoding is how to make estimated curve matches the real one for a certain target bitrate. Obviously, it needs only the logged qp in the 1st pass to be as closed to the one used in the 2nd pass as possible, while the real qp used in 1st pass does not need to be that restricted.
Because abr and crf use exact the same rc method, except for that abr adds a qp penalty system based on encoded bitrate cost. And 1st pass stats file logs only a rough qp-cost estimation near the encoded 1st pass' bitrate, so for example the stats file of a qp=19 slice may be identical to the same one of qp=20 in the 1st pass. So only if the bitrate or qp-penalty gives a too large gap will those two different rc modes in 1st pass have real effects.
multi-pass generally logs qp-bitrate curve in the early pass, and estimates the real qp-bitrate curve for the following pass, according to which the qp-offset is calculated to match the target bitrate. And if the estimation differs from real encoded one, a cost penalty will be added to the following frames to be encoded. So the problem in multi-pass encoding is how to make estimated curve matches the real one for a certain target bitrate. Obviously, it needs only the logged qp in the 1st pass to be as closed to the one used in the 2nd pass as possible, while the real qp used in 1st pass does not need to be that restricted.