6-pass encoding with DivX505Pro is an overkill?!!

[snowbeach]

Tested it with Gladiator, ca. 2:35h, 1 CD!
I used GKnot 0.28 with Lanczos Resize, 640x256, 494 kBit/s.
6-pass encoding with DivX505Pro was really an overkill and I have not seen any differences in picture quality to the 2-pass encoding!!
I doubt that the results will be better on 2 CDs. So, I am going to stay with 2-passes and about 8h encoding and not 6-passes and 24h of encoding.
Has somebody another experiences?

[jggimi]

Not that I can tell you anything about "multipass" and its plusses or minuses, but .... I'm betting that at such a low bitrate, you won't see much difference between passes.

You might take a single chapter and try several tests at several different bitrates, such as what Doom9 has done several times with Matrix and SPR for his careful MPEG-4 comparisons.

[snowbeach]

Thanks for the info. But I think I gonna stay with 2-passes until I will get a 6 GHz machine one day! :D

[Sirber]

2-pass is the best for every codecs.

[magomago]

would ~1600-2000 bitrate make a difference in 6 pass?

[madoka]

Pardon my ignorance, but I thought multi-pass should have a greater effect on low bit rate encodes. Since the first pass is done with high bit rate, low quantization, the farther away the (user selected) bit rate is from it, potentially the larger the deviation in size / quality would be. More passes would give the codec more chances to correct for the deviation. However, if the selected bit rate is high (therefore close to the first pass setting), then the deviation would be small and thus more passes would be unnecessary.

[N_F]

With 5.03 when the multipass feature was introduced it was quite clear that more than 2 passes was quite unnecesary. I haven't seen any documented changes that imply it should be different in 5.05.

So, yeah, 6-pass is overkill.

[bkam]

Originally posted by snowbeach
I doubt that the results will be better on 2 CDs.While I must admit I'm ignorant of multi-pass, I must disagree with the statement that an extra CD will not help how your encode looks. If you're unhappy with the one CD rip, you will almost undoubtedly be happier with the two CD rip.

[Acaila]

I believe snowbeach meant that the difference between 6 passes and 2 would be the same for either 1 or 2 CD's. I don't think he meant to say that 2 CD's don't give a better video than 1 CD.

[CaptainCarrot]

Hm, shouldn't it be possible to check the diffrence in bit distribution with that new ekg tool? Or maybe even by simply comparing the log files of the diffrent passes somehow else? If there is no (or hardly any) diffrence, you know that you're wasting cpu time.

[BiaTch 5.0]

I have compared my encodes & nth pass does lessen blur compared to old 2 pass. I think bitrate modulation may be the key to improving high motion scenes which I have found lately DivX suffers with.

[bkam]

Originally posted by Acaila
I believe snowbeach meant that the difference between 6 passes and 2 would be the same for either 1 or 2 CD's. I don't think he meant to say that 2 CD's don't give a better video than 1 CD. Ahh, you're probably right :). I apologize.

[magomago]

I just did End of Evangelion @ ~3030 Bitrate in two and four passesa and the 4th pass looks a LOT better(yeah - I'm a beleive of 30 minutes a CD when it comes to movies). Its as if everything is clearly defined and less muddled. I don't think I'd go higher since It'd seem useless but I'm pleased at the results

[CaptainCarrot]

Ok, it did some tests with the logfiles from diffrent passes. I encoded the first 33000 frames of MIB2 (first vob) and did 6 passes manually, 512x272, simple resize (for the speed), 800kbps and i made a backup of each logfile. Then i compared the bitrates spent for each single frame between two consecutive passes and calculated the average change/frame. This is what I got:

1st pass -> 2nd pass: mean change per frame of 19250 bit
2nd -> 3rd: 2040
3rd -> 4th: 872
4th -> 5th: 682
5th -> 6th: 599

When taking a closer look (well, not too close, it's 33000 frames after all) i found that in passes 4, 5 and 6 most changes are actually quite small, whereas few frames undergo extreme changes in bitrate. Some frames are even oszillating between two values from pass to pass. So there shouldn't be too much general improvement.
From the sheer numbers I'd say that the 3rd pass is the last one to gain a little bit of noticable quality, at least for this test clip.

[dTb]

From my personal experience without extensive testing I find that extra passes can help reduce macroblocking caused by motion. I generally do 4 passes because I have a speedy machine.

Some info that may interest people, I read a post by Gej on divx.com where he said bitrate modulation set at constant actually favours low motion slightly and that people may find a value of -0.01 or -0.02 to provide better overall quality.
Also, if you place a file in your c:\ drive called framelevelcontrol.txt you will have a lot of debug like info dumped there after each pass. The important info for me anyway is right at the end where there is a list of quantizers and how many frames used that quantizer. This can be handy to see how the quant distribution changes each pass. Be warned however that this file can be in the order of 100MB for a full movie, notepad has huge problems trying to open files this large.

[crusty]

The important info for me anyway is right at the end where there is a list of quantizers and how many frames used that quantizer

There's already a tool called DRFAnalyzer out there that will show you the number of frames per quantizer. try it out, it has a lot of useful info.

Personally I find EKG modulation much more useful than having a new pass just for it's own sake.
Just do two passes, then use the EKG tool to give the crappy scenes more bits, and do another pass. If you're not happy with 1 or 2 EKG modulated passes then you should probably give your movie more bits or lower the resolution.
Nth pass encoding probably was devised just for the sake of using EKG with it. It probably just was finished earlier than the EKG tool, so it was put into 5.03 and EKG came in the next version.

@magomago
30 minutes a CD? LOL ! :p
You should buy a DVD burner ASAP!

[magomago]

What is so funny with 30 minute a CD? And why a DVD Burner? I'll get one in about three years once they are much more affordable. Screw blueray for now mainly because I think its ridiculous that they are going to try to force a new format upon us. When I encode Series I put 2 episodes (each around 24 minutes) on a disc.

[dTb]

I might check out that program crusty but my interest in quant distribution over multiple passes has waned. The EKG is virtually useless for me because I just don't have the time to check a movie for scenes that I'm unhappy with after each pass. I can however easily do a couple extra passes and I find the increase in quality significant enough for me to warrant this.

I find it interesting with the importance people place in quantizers that they don't notice much if any improvement from extra passes. I know that quant distribution does change quite a bit with each pass (I assume the differences lessen with each pass but haven't really tested this assumption) so I tend to think maybe they aren't the be all and end all they are sometimes made out to be.

[BiaTch 5.0]

Originally posted by dTb
I might check out that program crusty but my interest in quant distribution over multiple passes has waned. The EKG is virtually useless for me because I just don't have the time to check a movie for scenes that I'm unhappy with after each pass. I can however easily do a couple extra passes and I find the increase in quality significant enough for me to warrant this.
Yes!!!