2nd pass with DivX 5.0 started...

[dragoman]

Hi,

Started the 2nd pass with the basic DivX 5.0 codec now...
Soft bicubic resize filter

Settings were same, 640x272 resolution, 1000kps bitrate

max quant 2, min 12

System is still 800 Thunderbird, 448 MB RAM

Even now it seems like the 2nd pass speed increase is true. My first pass results had an average fps of 12.

Right now I am sitting at around 16fps, which is a 25% increase in speed over the first pass. That's better than DivX 4.xx, since I never saw a difference speed-wise between 1st and 2nd pass with 4.xx.

I'll keep you updated....at 22% right now...

dragoman

[dragoman]

42% now....seems to be quite a bit faster, at least 15-16fps using full processing mode in Virtualdub...

if I used fast processing it would probably be faster, but I wanted the best quality.

We'll see in a couple hours...

dragoman

[alkasecond]

Mine's still at first 640x320 , ~30fps , P4@2.13GHz.
Not much different from 4.12.
But I have question , I set human(..whatever) enchancements at
normal.
Now I see people are reporting lots of shitty frames with that
setting.
Was it a mistake?
What do you think?

[dragoman]

Hi,

I set that at normal also...from what I've read it seems like it's a pretty cool thing.

We'll see what it does....50% now

dragoman

[dragoman]

Hi,

I've been looking at the output video in Virtual dub.....increasing it to like 4x normal size.....when I freeze the frame, it seems like the size of the blocks on walls and flat objects is much smaller.

The quality seems to be very good, if a bit blurry. The edges of objects are not quite as sharp, might benefit from a sharper resize filter than what I am using (soft bicubic).

60% now....

dragoman

[alkasecond]

62% into 2'nd pass , ~37-40fps that's ~20% faster than first.
It drops down to 35 in fast motion scenes and goes up to 42 in low.
Could it be that 40% faster they advertise was only for athlonXP?
What about P4?

[dragoman]

I hate you people with 2+ GHz pcs.....

What movie you encoding btw? I'm doing Independence Day Special Edition, mainly cause I bought it yesterday :) Still it's over 2 1/2 hrs long....long movie to encode.

Can't wait until my XP 2000+ arrives....

dragoman

[alkasecond]

Relax , your xp2000+ will probably do better than mine P4.
But in all the fairnes , why didn't they optimized it equally well
for P4 as well as for XP?

[alkasecond]

Doing AI , 600KBps , trying for 1cd rip.
Since that's my first rip (with 5.0 that is) it's hard to say
what size I'm gonna get.
Gave it a little more bitrate cause it supposedly compress better
than 4.12.

[dragoman]

Hey you are getting 50fps with an intel.....I think it is optimized enough!!!

dragoman

[alkasecond]

40 not 50 , but that's my goal.
Imagine doing rip watching the movie.
By the time movie's over you have rip ready(2 pas that is).
How that's sound?

[alkasecond]

Done , AI 2h25m second pass 1h30m.
I'll check quality tommorow.
Time to go.

BTW AI , 2pass , 640x320 ,600KBps , no audio = 661,474,132 bytes

[alkasecond]

Allright I added audio(128,cbr), cut all of credits and logo
still 759,869,440 bytes.
Damn have to reencode.
Bye.

[dragoman]

Done with 2nd pass, time was 4hr, 10min. Average fps for this encode was around 15-16 with an Athlon 800 w/ 448 mb ram

End file is 1.07 GB with settings:

1000kb/s bitrate
640x272 resolution

max quant 2, min quant 12

I am gonna wait to mux the audio with it.

I am going to try to use the new DivX 5.0 Pro codec now.

We'll see what a difference b-frames make and the other new options with Pro.

Here's a couple of screenshots of difficult to encode (I think scenes)

Image One
This is a shot of the plane silhoutted against the fire. Not bad, pretty good detail of the flame.

Image Two
Another shot of the flame rushing back through the plane and silhouetting the men against it. The edges of the flames are slightly blocky, but not overwhelmingly so.


The shot of the plane exploding against the background of the ship above. The sky, what little we see, is smooth, and the details of the planes themselves come through sharp against the background. Note the left-most plane's exhaust detail....looks very good.

Image Four
I chose this shot since it incorporates smoke and action. In the middle of the smoke there is some macroblocking, but it is small. The edges of the smoke are actually quite sharp. Another indication that DivX 5.0 is a step ahead of 4.xx

You can download these images in the attached zip file.

dragoman

[kxy]

How is the image compare to divx3.11 with nandub?

[ppera2]

Originally posted by alkasecond
62% into 2'nd pass , ~37-40fps that's ~20% faster than first.
It drops down to 35 in fast motion scenes and goes up to 42 in low.
Could it be that 40% faster they advertise was only for athlonXP?
What about P4?

When you encode MPEG2 source with resizing encoder itself have aprox. some 50 % CPU time, other is for IDCT and resize (plus frameserving).
So when you get 20 % better fps it means that encoding is faster about 40 %.

See my post 'DivX 5 first impression'. With Mpeg1 source without resize i have 60 % faster second pass. In this case encoder has about 90 % CPU time, I think.
I should make test with bigger size Mpeg1 source. Will encode something to it and see....

[muralin]

i just completed a rip of Pearl Harbor with DivX5. The result was outstanding. 2nd pass took a tad over 3 hrs (close to real time) and pass 1 took a bit over 4hrs. I am using a Dell 1.5GHz Win XP Pro machine.

cant complain.

[Acaila]

Lol! Real time! :D

It takes me 15 hours per pass to encode that monster (XviD) :(

[Helmut Hamburger]

lol^^^^^


try using psychovisual encoding set @ light combined with preprocessing @ light (when using nuetral bicubic resize)

its quite nice

normal psychovisual is too much and makes the video look bad

[alkasecond]

Originally posted by ppera2


When you encode MPEG2 source with resizing encoder itself have aprox. some 50 % CPU time, other is for IDCT and resize (plus frameserving).
So when you get 20 % better fps it means that encoding is faster about 40 %.

See my post 'DivX 5 first impression'. With Mpeg1 source without resize i have 60 % faster second pass. In this case encoder has about 90 % CPU time, I think.
I should make test with bigger size Mpeg1 source. Will encode something to it and see....

Thanks for bringing this up. Guess I wasn't thinking clearly
yesterday(pretty late).Agree 100%.

As for quality I must say I'm disapointed.
I've done few 2h+ 1cd rips before using 4.12 and in my (subjestive)
opinion they look better.
Maybe was that "psychimetric" thing (I set it to normal)?
In any case tommorow I'll try different movie , we'll see.
AI is jus not worth it(too touchy for my taste).

[dragoman]

Hi,

I see people have downloaded my screenshots...any comments about the quality? Was I right in what I said about them?

Any comments appreciated..

dragoman