x264 qcomp and ratetol

[TheBashar]

I was hoping someone could explain to me why the default value of ratetol is so low (1.0). Do I understand correctly that qcomp is responsible for "deciding" to what degree high motion scenes are compressed more than static scenes? If this is indeed true, why would we want to, by default, constrain bitrate variance to within 1.0?

After my searching on qcomp and ratetol it seemed to me like you would want a very high ratetol to let qcomp do its job. Have I misunderstood what these parameters are for?

[bond]

the higher the ratetol the more you allow the bitrate in 1pass to varry, but you also dont reach your target bitrate less closely
with a ratetol you should reach your target bitrate, but also allow the bitrate to vary

its always a tradeoff

[Sagittaire]

qcomp : QP curve compression
Ratecontrol equation : blurCplx^(1-qComp)

if qcomp = 1.00 then quantizer is constant for second pass. Real variable bitrate with constant quality.

if qcomp = 0.00 then bitrate is constant for second pass. Real constant bitrate with variable quality.

defaut value is 0.60 (best value for dev eyes)
best value for metric is in 0.70-0.80 interval

[TheBashar]

the higher the ratetol the more you allow the bitrate in 1pass to varry, but you also dont reach your target bitrate less closely
with a ratetol you should reach your target bitrate, but also allow the bitrate to vary

its always a tradeoff

So if I'm doing multipass, there seems like there would be less of a reason to keep ratetol low... right?

[TheBashar]

if qcomp = 1.00 then quantizer is constant for second pass. Real variable bitrate with constant quality.

Wouldn't it be "constant compression" as opposed to constant quality? I mean, isn't that the reason behind the curve anyway: the eye perceives less clearly in high motion scenes. I thought constant quality was the goal of the curve was to allocate less bits to high motion to maintain a more constant psycho-visual quality.

[chen]

qcomp : QP curve compression
Ratecontrol equation : blurCplx^(1-qComp)

What is QP curve? How is it compressed?

What is blurCplx? What does the value of the Ratecontrol equation mean?

[chen]

Wouldn't it be "constant compression" as opposed to constant quality? I mean, isn't that the reason behind the curve anyway: the eye perceives less clearly in high motion scenes. I thought constant quality was the goal of the curve was to allocate less bits to high motion to maintain a more constant psycho-visual quality.

It depends on how we understand the quality "constancy"? When we set qcomp to be 1.0, the quantisation parameter is a contant, thus the quantisation errors of different pictures are at the same level. In this sense, the quality is constant. On the other hand, if we set qcomp to be 0.6, the quality is said to be constant in terms of human visual perception, although different frames may have different levels of quantisation error due to different QPs.