Lower OPV Q-values w/2.70

[archaeo]

I've been noticing that I'm getting consistently lower Q-values on OPV runs when using CCE 2.70. I've tested this alongside CCE 2.50 by using the same source, same settings, and see that 2.70 has been delivering a Q that is at least a few points below that of 2.50. Just wondering if anyone else has noticed this.

[Sir Didymus]

Really ?

That's interesting... :)

Two possibilities exist:

1. in OPV mode, CCE 2.70 is intrinsecally capable of encoding providing better quality than 2.50...

2. The meaning, the scale, the values, or other factors affecting the Q parameter have changed, independently to the encoding quality...

As you know, looking at the funny story about of the quanti scale range [changes form 0..64 to 0..100 and back to 0..64] I would wait a little before assuming point 1 is true...

Cheers,
SD

[jdobbs]

I wouldn't worry too much about the number itself, it's probably just scaling (how many Qs can fit in a 10 lb bag?). Look instead at the quality you get in output and the nearness of the sizing. Those are controllable.

[archaeo]

it's probably just scaling (how many Qs can fit in a 10 lb bag?). 2. The meaning, the scale, the values, or other factors affecting the Q parameter have changed, independently to the encoding quality...


Yes that looks like the case jdobbs, sir didymus, just differences between how 2.70 & 2.50 estimates sectors... not any 'increase' in encode quality it appears. Looking at a particular sample, the predicted size between the two OPV runs varies, even at the same (Q=27) value:

CCE 2.70:
- Sampling 1992 of 198573 frames.
-- Predicted size (sectors) at Q=27: 1,029,328
-- Predicted size (sectors) at Q=10: 1,731,831
-- Predicted size (sectors) at Q=9: 1,798,441
-- Predicted size (sectors) at Q=8: 1,867,111
- Q Value selected: 8

CCE 2.50:
- Sampling 1992 of 198573 frames.
-- Predicted size (sectors) at Q=27: 1,090,592
-- Predicted size (sectors) at Q=11: 1,942,711
- Q Value selected: 12

The good news too is that I am getting OPV sizes falling pretty consistently between 4.1 and 4.3 Gb (w/RB v72)
edit: only with higher Q results

[archaeo]

jdobbs wrote:
Look instead at the quality you get in output and the nearness of the sizing

What I am seeing is some undersizing based upon the lower Q-values generated by CCE 2.70. It seems to happen more when the resultant Q is low, as in single digits... As a test, I ran the same source w/RB v73 through both 2.70 and 2.50:

CCE 2.70 final size=3.3Gb (Q=1)
CCE 2.50 final size=4.2Gb (Q=2)

I guess I'll just have to live with the difference in Q-value, and therefore OPV sizing discrepancies?

[robot1]

Originally posted by archaeo
jdobbs wrote:


What I am seeing is some undersizing based upon the lower Q-values generated by CCE 2.70. It seems to happen more when the resultant Q is low, as in single digits... As a test, I ran the same source w/RB v73 through both 2.70 and 2.50:

CCE 2.70 final size=3.3Gb (Q=1)
CCE 2.50 final size=4.2Gb (Q=2)

I guess I'll just have to live with the difference in Q-value, and therefore OPV sizing discrepancies? CCE can't do better than Q=1, from what I can understand.
So, you have already achieved maximum quality.

[archaeo]

yes, it was a high bitrate source (ABR=4885), but even with compression at 65% it came out w/Q-value of 1! Tried to tweak through RB-Opt with different Q, larger sizing, but as you know, RB-Opt isn't quite ready to work with CCE 2.70. So I just took it and ran through 2.50, with acceptable sizing results.