Log in

View Full Version : Baseline CQ not constant quality


fields_g
29th March 2009, 20:41
I tried out the new MeGUI preset for the G1. After switching from 2 pass to CRF 18, I noticed that quality (especially dark/noisy areas) was good at the beginning of the GOP and got worse throughout and pulsed to great quality at the beginning of the next GOP. Thinking that it might have something to do with the VBV, I switched back to the vanilla G1 preset. Same problem still persisted. I also tried changing the min gop size to 1 and still no joy. With the little bit of tinkering I've done so far, the best things to do is to denoise and reduce the keyframe interval (more less noticeable pulses).

Avisynth Script:
DGDecode_mpeg2source("VTS_02_1-cut.d2v", info=3)
ColorMatrix(hints=true, interlaced=true, threads=0)
tfm(order=1).tdecimate(hybrid=1)
#crop
Spline36Resize(448,320) # Spline36 (Neutral)


Vanilla G1 Commandline
--pass 2 --bitrate 1000 --stats "VTS_02_1-g1.stats" --level 3 --deblock 1:0 --no-cabac --subme 8 --partitions p8x8,b8x8,i4x4 --qpmin 12 --vbv-bufsize 2500 --vbv-maxrate 2500 --me umh --threads auto --thread-input --sar 85443:87500 --progress --no-psnr --no-ssim --output "VTS_02_1-g1-cut.mp4" "VTS_02_1-g1.avs"

I uploaded 30 seconds of the source and my encode here (https://netfiles.umn.edu/users/fiel0097/shared/G1/). Got any more recommendations?

ajp_anton
29th March 2009, 22:46
Extremely low VBV limits + crf... and you expect nice results?

fields_g
30th March 2009, 03:15
Extremely low VBV limits + crf... and you expect nice results?

Like I said, 2-pass is the same result. I just tried using the same G1 profile, setting the VBV limitations to 0 (automatic) and setting it for 10 Mbit 2-pass, I still get the same result. Other suggestions?

ajp_anton
30th March 2009, 04:33
Read the far east parts of your post very quickly as you made it very wide.
The problems appeared when you switched to crf, right? Are you saying that the settings that did work doesn't anymore?

Sharktooth
30th March 2009, 12:31
2pass CANT be the same result.
btw, try disabling psy-rd/psy-trellis