RV9 EHQ mode quality and PSNR ...

[Sagittaire]

Very surprising ... doesn't it

first pass: Iframe and Pframe quant4, bframe quant 8 (for quality 90)and EHQ=70 ---> PSNR = 51.39

second pass: Iframe and Pframe quant4, bframe quant 8 (for quality 90) and EHQ=100 ---> PSNR = 51.29


=======================================================
mes: 64
frd: 1
rda: 1
rdai: 1
hfe: 1
cpu_scal: 70
met: H
TR Q am PSNR Bits MiPSNR AvPSNR AvBits Fr.type
=======================================================
40 4 10896 99.99 4952 99.99 0.00 4952 Key 70
200 4 16 99.99 88 99.99 0.00 2520 70
80 8 0 99.99 64 99.99 0.00 1701 d 70
120 8 0 99.99 64 99.99 0.00 1292 d 70
160 8 0 99.99 64 99.99 0.00 1046 d 70
360 4 0 99.99 64 99.99 0.00 882 70
240 8 0 99.99 64 99.99 0.00 765 d 70
280 8 0 99.99 64 99.99 0.00 678 d 70

..........................................................................

60160 8 0 78.72 64 40.90 51.26 40227 d 70
60360 4 0 78.72 64 40.90 51.28 40200 70
60240 8 0 78.72 64 40.90 51.30 40174 d 70
60280 8 0 78.72 64 40.90 51.32 40147 d 70
60320 8 0 78.72 64 40.90 51.33 40121 d 70
60400 4 0 78.72 64 40.90 51.35 40094 70
60440 4 0 78.72 64 40.90 51.37 40068 70
60480 4 0 78.72 64 40.90 51.39 40041 70
=======================================================
mes: 64
frd: 1
rda: 1
rdai: 1
hfe: 1
cpu_scal: 100
met: F
TR Q am PSNR Bits MiPSNR AvPSNR AvBits Fr.type
=======================================================
40 4 10896 99.99 4952 99.99 0.00 4952 Key 100
200 4 16 99.99 88 99.99 0.00 2520 100
80 8 0 99.99 64 99.99 0.00 1701 d 100
120 8 0 99.99 64 99.99 0.00 1292 d 100
160 8 0 99.99 64 99.99 0.00 1046 d 100
360 4 0 99.99 64 99.99 0.00 882 100

...........................................................................

60240 8 0 83.73 64 41.47 51.19 40953 d 100
60280 8 0 83.73 64 41.47 51.21 40926 d 100
60320 8 0 83.73 64 41.47 51.23 40899 d 100
60400 4 0 83.73 64 41.47 51.25 40872 100
60440 4 0 83.73 64 41.47 51.27 40845 100
60480 4 0 83.73 64 41.47 51.29 40818 100

it's a very short sample but result is strange ...

[Sirber]

From Eiffel 65:
Too much EHQ will bring you underground:D

[karl_lillevold]

yes, a little odd, but things are weird when PSNR values are so high. A difference for one pixel makes a big difference. I have seen PSNR go down when lowering QP, for frames with almost no activity. It really makes sense to cap PSNR to perhaps 55-60, when calculating average or overall PSNR. And also, you don't know the filesize for the 1st pass, even though PSNR could be higher, the resulting (potential) filesize was maybe higher too, but this was never saved. What if you run 1 pass and compare 70 to 100? then you can see the filesize as well..