xvid 1.0 beta2 and fades in/out

[jkwarras]

Hi,

I've noticed that in XviD 1.0 beta2 teh codec handle quite bad fades to black. I've used 2 pass, MPEG quant, adaptative quant, trellis, VHQ1, bframes 3 (default settings offset and others), chroma optimizer, GMC. Average bitrate target 785 kbps. The results look great, but fades doesn't look good at all, a lot of blocking appears. It happens with and without filtering with avisynth (if a directly open the DV avi file in VDmod and encode it). I didn't see it happens with Xvid since a long time. I don't know if this is known, if someone notices it or is just me and my settings.

Anyway, just to make sure, i've done this test (no filtering with avisynth, directly opening video and ecoding in VDmod):

1) 2 pass averaging 785 kbps JUST the "'problematic" fade to black.
2) Run again a full (entire movie) 2 pass averaging 785 kbps.

Results: For the same fade, 1 looks better, 2 look blocky.

I'll try to post the clips later, cause i'm at work right now :)

Regards

[vijv]

i noticed the same problem but it was when it was fading from black to a more bright picture. this was beta 1 not 2, havn't tested 2. it didn't happen with the old koepi release

[Koepi]

Strange, for me it's quite the opposite, beta2 handles fades better than dev-api-3 builds. But then i use vhq=4 (maybe that's the solution to "bad" fades?).

Regards
Koepi

[vijv]

nope i used vhq 4, trelis, qpel, motion search precision 6. I forgot to mention that as it fades to a brighter picture, the picture get sharper because it starts off blurred. The picture is extremely blocky but post processing with ffdshow takes care of it.

[jkwarras]

Originally posted by jkwarras

1) 2 pass averaging 785 kbps JUST the "'problematic" fade to black.
2) Run again a full (entire movie) 2 pass averaging 785 kbps.

Results: For the same fade, 1 looks better, 2 look blocky.


Here are the clips. Just for the "problematic" fade.

1) http://www.arkania.org/jkwarras/video/01_2pass_just_fade.avi
2) http://www.arkania.org/jkwarras/video/01_2pass_full_movie.avi

To download "Save link as" othewise it doesn't work.

Sorry a mistake: this is a fade from to black to a bright image.

Difference is quite obvious, as for 1) the avg bitrate is 963 kbps and for 2) is only 267 kbps! (according to media info) but settings were exactly the same fot both encodings. I think that 2 pass is doing something worst when allocating bitrate to this fade. When you only encode the fad runnning a 2 pass, it looks better.

[Koepi]

As mentioned on other threads, try setting min/max overflow increase/degradation with 20% instead the default of 60% - this might help.

Regards
Koepi

[jkwarras]

Originally posted by Koepi
As mentioned on other threads, try setting min/max overflow increase/degradation with 20% instead the default of 60% - this might help.

Regards
Koepi

I'm trying this right now, i'll post resulting clip. Thanks for the tip, and thanks for this codec! :rolleyes:

[jkwarras]

Ok, did test the tip from Koepi. The results look better, but not as good as the 1). But it helps a lot since bitrate is almost the same as 2) and look a lot less blocky.

Here are the clips. Just for the "problematic" fade.

1) http://www.arkania.org/jkwarras/video/01_2pass_just_fade.avi
2) http://www.arkania.org/jkwarras/video/02_2pass_full_movie.avi
3) http://www.arkania.org/jkwarras/video/03_2pass_full_movie.avi

To download "Save link as" othewise it doesn't work.

[jkwarras]

Another try. Now with vhq4. Look better than 3) and 2) but not as good as 1). I'll try to resume :)

XviD Beta 2
-----------

For all encodings: 2 pass at target bitrate 785 kbps.
To download "Save link as" othewise it doesn't work.

1) http://www.arkania.org/jkwarras/video/01_2pass_just_fade.avi

(2 pass ONLY for the fade in question)

MPEG quant
Adaptative quantization
GMC
BVOPs: Max 3; Quant ratio 1.50, Quant offset 1.00, packed bitstream, Closed GOV
Max/MIn overflow%/degradation%:60% (default)
Motion search precision:6
Chroma motion
VHQ1
Trellis quant
Chroma optimizer
Rest=default

Final size = 588 KB
Avg bitrate = 799 kbps

2) http://www.arkania.org/jkwarras/video/02_2pass_full_movie.avi

(2 pass for full clip)

MPEG quant
Adaptative quantization
GMC
BVOPs: Max 3; Quant ratio 1.50, Quant offset 1.00, packed bitstream, Closed GOV
Max/MIn overflow%/degradation%:60% (default)
Motion search precision:6
Chroma motion
VHQ1
Trellis quant
Chroma optimizer
Rest=default

Final size = 326 KB
Avg bitrate = 256 kbps

3) http://www.arkania.org/jkwarras/video/03_2pass_full_movie.avi

(2 pass for full clip)

MPEG quant
Adaptative quantization
GMC
BVOPs: Max 3; Quant ratio 1.50, Quant offset 1.00, packed bitstream, Closed GOV
Max/MIn overflow%/degradation%:20% (koepi suggestion)
Motion search precision:6
Chroma motion
VHQ1
Trellis quant
Chroma optimizer
Rest=default

Final size = 318 KB
Avg bitrate = 249 kbps

4) http://www.arkania.org/jkwarras/video/03_2pass_full_movie_vhq4.avi

(2 pass for full clip)

MPEG quant
Adaptative quantization
GMC
BVOPs: Max 3; Quant ratio 1.50, Quant offset 1.00, packed bitstream, Closed GOV
Max/MIn overflow%/degradation%:20% (koepi suggestion)
Motion search precision:6
Chroma motion
VHQ4 (maybe help fades)
Trellis quant
Chroma optimizer
Rest=default

Final size = 318 KB
Avg bitrate = 250 kbps

CONCLUSION
----------

Personal and subjective ranking:
1º - (1) 2 pass only for the fade in question
2º - (4) vhq4 + max/min overflow/degradation to 20%
3º - (3) max/min overflow/degradation to 20%
4º - (2) settings as (1) but encoding entire movie

SO, vhq4 and max/min degradation/overflow to 20& help fades BUT it's quite strange that runnign a 2 pass only for the fade turns a better looking file that the entire movie for exactly the same settings. The overall bitrate distribution doesn't seems to give suffisent bitrate to this situations. And even if vhq4 and max/min settings help, the same situation arise: 2 pass only for the fade is far better than for entire movie. Just look bitrates, in (1) is just a little oversized (target is 785 and result is 799), and in other cases (2) to (4) when running a full encodign the bitrate in this part is just too low for handling this well (250 kbps).

However, this is just a personal conclusion. :rolleyes:

[Koepi]

Just to get more insights what you're doing:

- how long is "the fade only scene" which has 4 times the bitrate than the other scenario?
- how long is the "total clip"?

The bitrate distribution algorithms are most efficient on "real movie encodes" in the terms of length (i.e. 30 minutes and more).
For trailers they tend to do terrible things sometimes, but then, it's a matter of tweaking which is possible through the vfw gui. If you choose a real hard scenario without really knowing what you're doing you won't be pleased with the result - which i think is only a natural phenomenon.

(No offense in that, be assured.)

Regards
Koepi

[jkwarras]

Originally posted by Koepi
[B]Just to get more insights what you're doing:

- how long is "the fade only scene" which has 4 times the bitrate than the other scenario?
- how long is the "total clip"?

- "Fade only scene": 150 frames
- "Total clip": 2 min. (is a shortfilm)

The "fade only scene" that I post is a little shorter (in length) just because when i cut from frame 100 to 250 in the full clip encoding (scenario 2 to 4) it doesn't cut exactly between those frames.

The bitrate distribution algorithms are most efficient on "real movie encodes" in the terms of length (i.e. 30 minutes and more).
For trailers they tend to do terrible things sometimes, but then, it's a matter of tweaking which is possible through the vfw gui.

That makes sens. Maybe the clip is too short for allowing the algorithm to be efficient.


If you choose a real hard scenario without really knowing what you're doing you won't be pleased with the result - which i think is only a natural phenomenon.

Am i doing something stupid? :p My settings seems quite standard to me.

(No offense in that, be assured.)

No offense at all. Just trying to report something I've found, maybe it will help, maybe not. Excellent job, XviD is very, very good, I love it :)

Regards

[vijv]

yes it was a trailer clip of matrix revolutions at the end of the deloaded dvd. Maybe if i did the full movie it would look a lot better

[Sigmatador]

I post for Karma Koma who have some problems with his forum account.

hi,
tried to encode The "Sting - All this time" DVD yesterday.


Encoded with Koepi's XviD24062003-1.exe with the following parameters :

BFrames : 2/150/100/50
VHQ : 4
MotionSearch : 6
Matrix : HVS Best picture
Chroma motion
Discard first pass : checked
All other options set to default (QPel disabled, etc.)


The 1st pass ended with a 994MB file, which gives 46% compressiblity.




Now, encode with XviD 1.0 beta 2
Same settings as with XviD24062003-1.exe but with GMC, Adaptive Quantization and Treillis Quantization checked (since they are told to be fully fonctional now).

The 1st pass ended with a 975MB file, which gives 47% compressiblity.



Here are the 2pass encode results
(the 2nd video starts 20 frames after vid1 due to different keyframes, sorry for that)

XviD 24062003-1 : http://membres.lycos.fr/rocksane/Vid1.avi
XviD 1.0b2Aloha : http://membres.lycos.fr/rocksane/Vid2.avi



The XviD 1.0b2 video looks ugly to me. :(
I'm not talking about mosquito noise and pixelation ; that has to do with bad compressiblity (Anyway, it might be interesting to compare both vids).
I'm talking about fades, and color bugs on Sting's face at the end (0'06 and after) , this looks like DivX 3.11 bugs. The worst thing is the moving background (0'03->0'07, when you see the audience). It shouldn't move!


Do you have the same problem, or do I have a decoder/version/anything else problem ?
What do you think about that ugly video ? (especially the background)



NB1 : I use XviD decoder to decode, not ffdshow.
NB2 : also encoded in DivX 5.1.1 3pass @ 870kbps (same filesize), with Slowest Performance/Quality ratio and BM set to 0.2. The result was very close to XviD. Less mosquito noise but less sharpness :( I prefer XviD (a little)
NB3 : I post my Avisynth script, maybe it can help :
mpeg2source("sting.d2v")
FieldDeinterlace(blend=false)
undot()
crop(10,0,698,576)
deen("c3d",1,1,2,1,2,0.5,9)
BicubicResize(608,352,0,0.7)
limiter()

[Manao]

Using FFDshow to playback, the result was very good for the XviD 1.0b2. No problem whatsoever. It was clearly better than the clip made with the previous version. Nothing wrong in the background, no color distorsion ( that's ab iDCT bug, btw, so if you get such an artifact, change the iDCT method in FFDShow ), almost no ringing / mosquito noise.

I tried to playback it with XviD Dshow, the result was worse, but not that much. Anyway, I can't spot the artifacts you're talking about. And I don't know how you can playback with XviD DShow the first clip, since I can't ( which is normal ). So I'm lead to think you've got a strange decoder. May be Nic's one ?

[jkwarras]

I also don't see any artifacts. Video look great, and XviD beta2 looks a little better (sharper). I'm using ffdshow to decode it.

[yaz]

@jkwarras, manao, sigmatador & all :
when u mention ffdshow, what version do u mean ? i've tested the last 4 releases from athos. although they were different, any of them gave me better output than xvid.ax. further, the best output i got with mpc wout ffdshow &(!) xvid. see details here (http://forum.doom9.org/showthread.php?s=&threadid=66507)

the bests
y

[jkwarras]

Originally posted by yaz
@jkwarras, manao, sigmatador & all :
when u mention ffdshow, what version do u mean ? i've tested the last 4 releases from athos. although they were different, any of them gave me better output than xvid.ax. further, the best output i got with mpc wout ffdshow &(!) xvid. see details here (http://forum.doom9.org/showthread.php?s=&threadid=66507)

the bests
y

I'm using ffdshow from 16/08/2003.

[vijv]

is it recommended so set the min/max overflow increase/degradation with 60% or 20% for a full movie encode (Terminator 3)in beta 2?

[jkwarras]

Originally posted by vijv
is it recommended so set the min/max overflow increase/degradation with 60% or 20% for a full movie encode (Terminator 3)in beta 2?

I don't really know. Apparently it was talk in another discussion around this forum but as always: try it :)

If i've understand it correctly it will decrease the speed of the codec reaction, so it will avoid some artifacts between low/high motion scenes. Someone correct me if i'm wrong. try it with vhq4: slow, but impresive results.

[vijv]

i did the movie with 60 with qpel and vhq4 and i don't see any of the blockiness that i saw when i did the movie clip.
I am not going to do 20 as i am extremely pleased with the result. it is almost like watching the dvd when pp is used with the new decoder. however my computer is using 50% of it processing power to decode, so i don't think my xbox will be able to cope.

[Teegedeck]

The price for lowering the overflow values could be less efficiency and a slight overall decrease in quality, I would guess.

[sysKin]

Originally posted by Teegedeck
The price for lowering the overflow values could be less efficiency and a slight overall decrease in quality, I would guess. I'd say that lowering overflow values *increases* quality and improves quant distribution, at the cost of possibly making codec miss desired filesize.

Radek

PS. Newest 2-pass experiments show much better quantizer distribution, especially in things like fades. Stay tuned for beta3.

[Teegedeck]

bummer. Sorry; I had thought that the overflow would have to be compensated for on a larger number of frames and a slight degradation could be visible. Thanks for correcting that.