Log in

View Full Version : Should I do an extra pass?


DJ Alik
14th September 2003, 01:57
Hi. I am wondering whether I should do an extra pass. How much more can the quality of a file increase with more passes?

I am encoding "The Devil's Advocate" on to 2 cds DivX5.1 Standard B frames and GMC no qpel Psych slow 1262kbps mv enabled on all passes
704x288

So far I made 4 passes and here are the stats that I got using the DivX DRF analyzer ver 0.951 :

Pass 2:
DRF=1&2: 5059 2.6%
DRF=3: 93023 47.9%
DRF=4: 91688 47.2%
DRF=5: 0 0.0%
DRF=6: 4545 2.3%
DRF=7: 0 0.0%
DRF=8: 0 0.0%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 1.80%
KeyDRF<4: 3376
KeyDRF=4: 110
KeyDRF>4: 20
AverageKeyDRF: 2.59 MAXDRF: 7 AverageDRF: 3.50 Deviation: 0.67

Pass 3:
DRF=1&2: 3393 1.7%
DRF=3: 97152 50.0%
DRF=4: 93208 48.0%
DRF=5: 0 0.0%
DRF=6: 562 0.3%
DRF=7: 0 0.0%
DRF=8: 0 0.0%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 1.80%
KeyDRF<4: 3395
KeyDRF=4: 109
KeyDRF>4: 2
AverageKeyDRF: 2.48 MAXDRF: 6 AverageDRF: 3.45 Deviation: 0.56

Pass 4:
DRF=1&2: 3116 1.6%
DRF=3: 97352 50.1%
DRF=4: 93637 48.2%
DRF=5: 0 0.0%
DRF=6: 210 0.1%
DRF=7: 0 0.0%
DRF=8: 0 0.0%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 1.80%
KeyDRF<4: 3433
KeyDRF=4: 56
KeyDRF>4: 17
AverageKeyDRF: 2.43 MAXDRF: 6 AverageDRF: 3.45 Deviation: 0.55

I see that with each pass there are more frames with quant3-4 and less of frames with quant1,2,6. Is that a good thing? Any feedback would be very appreciated

twist3d
14th September 2003, 14:01
[sarcastic mode on]
should i wear today:

(a) red shirt
(b) blue shirt
(c) green shirt
(d) no shirt at all

[sarcastic mode off]

But seriously, I don't think that forum polls were meant for this generally (I could be wrong).
Trust on your eyes on the 2-pass, 3-pass etc. results.

Just my 2c.

SeeMoreDigital
14th September 2003, 18:17
Agreed,

If you're really after getting the best quality possible then multipasses may not be the answer for you!

Have you had a go yet with applications such a DivX's EKG?

And why 704x288? I would have thought 720x304 would have been nearer the mark!

Cheers

DJ Alik
16th September 2003, 05:51
So I decided to another pass but changed the performance quality to slowest and disabled the mv file. Everything else remained the same.

And 28hours 42 minutes later :scared: here is what i got.

Pass 5:
DRF=1&2: 5802 3.0%
DRF=3: 96863 49.8%
DRF=4: 91650 47.2%
DRF=5: 0 0.0%
DRF=6: 0 0.0%
DRF=7: 0 0.0%
DRF=8: 0 0.0%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 1.80%
KeyDRF<4: 3465
KeyDRF=4: 41
KeyDRF>4: 0
AverageKeyDRF: 2.36 MAXDRF: 4 AverageDRF: 3.42 Deviation: 0.57

So I was like whooo who. The quality should be awesome if you look at the stats. But after examining the movie I am not very happy :mad: There are tons of microblocks especially on the walls and faces. There is also a lot of ringing on the edges. I thought maybe it's the post processing so I set demacroblocking and deringing to maximum but that didn't help much. So I don't know what to do.

I am gonna try to post a few frames in a little bit.

SeeMoreDigital
16th September 2003, 10:54
But after examining the movie I am not very happy There are tons of microblocks especially on the walls and faces. There is also a lot of ringing on the edges. I thought maybe it's the post processing so I set demacroblocking and deringing to maximum but that didn't help much. So I don't know what to do. I know what you should do..... don't bother doing it! I can't think of anything more upsetting or disheartening spending 28 hours generating an encode that your not happy with!

Unfortunately it's far too easy to criticize an encode, even your own. And different viewers will always see different flaws in the same image anyway!

There really are too many variables!

But that's just my opinion and I don't want to rain on your parade!

Cheers

DJ Alik
17th September 2003, 04:02
I think the reason it looks that way is because the movie is kind of old and when you look very closely at the source it does look very grainy. I have never messed around with any denoising filters so i don't know what to do.

DJ Alik
17th September 2003, 08:32
OK here are a couple of frames.

Here is the original frame from the source as you can see it's very grainy (you can easily see the grain if you zoom to like 200%)

http://djalik1.narod.ru/original.tif

and here is the encoded frame.


http://djalik1.narod.ru/encoded.tif

can anyone suggest me what's the best way to remove the grain from the movie?

DJ Alik
18th September 2003, 23:31
Well I decided to start over :(

I generated two more passes using convolution3d("movielq") and bilinnear soft resizing.

It helped a little bit but there is still flickering going on. When I examined it frame by frame you could see how the grain appears and disappears. I noticed that the key frames are the most grainiest. Can anyone suggest anything else on how to remove the grain without losing the quality?

Waiting to hear for replies

cordraconis
21st September 2003, 14:36
Originally posted by DJ Alik
Well I decided to start over :(

I generated two more passes using convolution3d("movielq") and bilinnear soft resizing.

It helped a little bit but there is still flickering going on. When I examined it frame by frame you could see how the grain appears and disappears. I noticed that the key frames are the most grainiest. Can anyone suggest anything else on how to remove the grain without losing the quality?

Waiting to hear for replies
Wel, it's a matter of opinion, but i'd use (neutral)bicubic in stead of bilineair for the resizeing. Something in the back of my head says that bilineair is too harsh. (dunno why I have this feeling.)

The flickering sounds odd to me, especially if you say that the grain alternates.:confused:
Therefore I'd recommend some sort of temporal smoothing, so the grainy frames are reduced by the other non-grainy frames.
Use the lowest setting of course.

If you want to try convolution3D, then take the numbers for movieLowQ, and double them. I used this with very good result for Alien 1.
When you try the different settings for the filters, do only a one-pass maximum quality encode, and note the filesize. use the filesizes (compressibility tests) as your guidelines.
As mentioned before, spending over a day for one pass, is not worth it, especially when trying filter settings. (trust me on this one, when i use filters, this is the way i go ;) )

oh, and the key frames are like huge photographs of a whole frame, so for me it is normal that they are more detailled (ie more grain) than the rest.

good luck!

(P.S. If you have a lot of hard disk space to spare, make a losless (huffyuv) file after using all the filters, and then use this pre-filtered file for your final multipass divx.

and also: maybe you could still lover the resolution to 640*xxx, if everything else fails.)

Cheers!
Hope this helps ...