More quality for dark scenes

[AsTimeGoesBy]

Sorry to jump on that point once more, but all the advices i have found aren't really satisfying...

Is there an established 'darkness-friendly' custom matrix in meantime or any other procedure to maintain details in dark scenes?

I just have started to re-encode the 'Lord of the Ring' set with Xvid 1.1b:
Profile @ level : AS @ L5
Encoding type / bitrate: 2 pass, 1900 kbps
Resolution: 720x312, 25fps
Consecutive B-frames : max. 2
Intra-frame tuning : 10/1/20
Overflow treatment : 5/5/5
Curve compression : 0/0/0
Adaptive quantization : disabled
Packed bitstream : enabled
motion search precision : 6 - ultra high
VHQ mode : 4 - Wide search
VHQ for B-frames : enabled
Use chroma motion : enabled
Quantizier restricion : I:1/31 P:1/31 B:1/31
Trellis quantisation : enabled
Zone settings: Chroma Optimizer
BVOP-Sensitivity : 0

Compared to Xvid 1.0x (with adaptive quantization enabled) it's a bit better now. However, while bright scenes are really DVD-authentic, some dark scenes still are too smooth and have lost many details. :(


...or will it help to make the 1st pass with more brightness and remove that brightness again before starting the 2nd pass?
(maybe a bit unsophisticated!?)

[Didée]

No ... we are still waiting for syskin's HVS-plugin-system-thingy :)

Current possibilities are:

a) Remove noise thoroughly, but carefully (especially in dark scenes noise is problematic - at least for my taste)

b) use CC to raise cheap scenes' bitrates - works good in combination with a)

c) go the hard way with individual zones ...

[AsTimeGoesBy]

Thanks for the fast answer!

to a) There is not much noise generally so my denoising settings are low.

to b) You mean this curve compression modification for low bitrates!?
But i don't know if dark scenes full of details are really 'low bitrate', however i can try it.
Must i redo the first pass? I still have the log file out of the 1st pass...

to c) Zones... the LotR films are switching a lot from dark to bright and back. This will be a lot of work.... :(

[Sharktooth]

b) you have to redo the first pass...

[Didée]

Originally posted by Sharktooth
b) you have to redo the first pass...
What? Why? Last time I checked, Curve Compression was a 2nd-pass feature, independant of the 1st pass.

Originally posted by AsTimeGoesBy
...or will it help to make the 1st pass with more brightness and remove that brightness again before starting the 2nd pass?
If at all, then the other way round - 1st pass darker, 2nd pass normal. You had to somehow make XviD *think* during 1st-pass that dark scenes are very-very cheap, Then on 2nd pass, XviD will assume those scenes could get small quantizers, because of the small 1st-pass stats.

However, that's playing bad tricks to the bitrate control system. Effectively the dark scenes will eat *much* more bitrate than XviD thought they *would* eat, and will start to compensate for that. Therefore, one would need to use very weak overflow control settings, like 1-5-5 or so. And by that, you'd totally loose size prediction, and would have to start with a lowered target size from the beginning (blind guess: 80%~85% of what you're really aiming for.)

Mind you, I've never tried to do it this way. It's "theory".

[AsTimeGoesBy]

about c)...
The other way around? - 1st pass darker, 2nd pass normal...?
However if i understand right the consequences for quantizier, i should lower the difference between dark and bright scene in the 'pseudo' 1st pass so that the quntiziers for dark and bright scenes are more similar. Maybe any gamma filter like levels() can do this...!?

I have no doubts that such a tweaking of Xvids bitrate investment may have inpredictable (for me) consequences, but at least the matter of size prediction isn't really important, i just like to see my Xvid around 2GB (+/-500mb), somthing like a 3:1 or 2:1 reduction of the DVD.

[Sharktooth]

Originally posted by Didée
What? Why? Last time I checked, Curve Compression was a 2nd-pass feature, independant of the 1st pass.
i must have smoked something really heavy before replying...

[pwh04]

Hi, I seem to have good luck using hvs best in combination with:

profile: as @ l5
one pass
quantizer: 2-4, 2-4, 2-4
trellis
targe bitrate: in range of 1200- 1400
adaptive quant
vhq for frames
vhq: 2
motion search: 6
bvops: default
chroma motion
chroma optimizer
resolution: 528/ mod 16
29.970 fps
ntsc 4:3
crop: usually avg 4-4-4-4

I definitely notice greater sharpness and brightness with these settings using hvs best. I also have had success with file prediction using this method

Hope it helps

[pwh04]

forgot to mention, I use:

deinterlace: tomsmo
resize: lanczos
Avs filters: none

[Andrey]

>>resize: lanczos
If you have ringing problem too then replace lanczos with bilinear...
Lanczos resize produce output that is sharper than original and too sharp edges can be a problem on mid/low bitrates...

[pwh04]

Originally posted by Andrey
>>resize: lanczos
If you have ringing problem too then replace lanczos with bilinear...
Lanczos resize produce output that is sharper than original and too sharp edges can be a problem on mid/low bitrates...

---------
Thanks for the tip!.. i have noticed this but couldn't figure out why only on occasion.

[AsTimeGoesBy]

Originally posted by AsTimeGoesBy
...or will it help to make the 1st pass with more brightness and remove that brightness again before starting the 2nd pass?
(maybe a bit unsophisticated!?)
Originally posted by Didée
If at all, then the other way round - 1st pass darker, 2nd pass normal. You had to somehow make XviD *think* during 1st-pass that dark scenes are very-very cheap, Then on 2nd pass, XviD will assume those scenes could get small quantizers, because of the small 1st-pass stats.


Well, that 'hint' don't work...
The file after the 2nd pass looks anything between Picasso and Andy Warhole...

[Soulhunter]

Maybe raise the brightness b4 encoding and lower it while playback... ^^


Bye

[AsTimeGoesBy]

Yes, encoding at a(ny) good brightness level is reasonable or even needed when using adapitve quantization, but 'optimal' brightness at the end is a matter of playback device.