1-pass Quality Mode V's 2-Pass

[DaveQB]

whats the pros and cons here??

i have always used the 2-pass method for anything important and worthwhile keeping.

but have recently been told that the 1-pass Quality Mode is better.

that it basicly has a greater range on each frame; by that i mean, uses alot more bitson frames that needed it and alot less on frames that dont.

i am going to do some experiementing with this 1-pass Quality Mode and see what i get,
was just wondering though what other people have found, things i may not have thought of or realised.

thanx for any feedback
:D

[Dark-Cracker]

hi,

the main drawback in the quality mode it u can rush the desired filesize (because it 1 pass, and u enter a quantizer instead of a average bitrate.)

the only way to use the 1pass quality (i speak for me) it if your 2 pass encoding was undersized . in fact when u make 2 pass , the 1 pass was in quant 2 (max quality), u can open the .sats file with gknot and it will show you the estimated final filesize, it this size was less thant the wanted size , u can make a 1 pass quality 100% (quant2) . u can try to make a compressibility test and if the result was upper than 100% use 1pass quality 100 %.

Bye.

[yingx2]

Glad that you brought this topic up as it's really something which needs to be discussed more.
I've been believing that constant quant encoding is the better one when it comes to picture quality for a long time, and now I believe it even more.

I found a thread where temporance wrote:

To simplify greatly, VBR rate control has one aim - to discover a constant quantizer that, when applied to every frame, will hit the desired filesize. Xvid does not achieve a constant quantiser for the 2nd pass, it varies a lot.

http://forum.doom9.org/showthread.php?s=&threadid=44200

It this holds true, I would assume that multi-pass enconding that people keep using nowadays is obviously inferior and more time-consuming than fixed quant enconding, because no matter how many passes you run, the best you can expect is that your encoding has reached a perfectly stable status, where every frame is being eoncded at The Same Quantizer -- whcih is exactly what 1-pass QB mode does for you. So why multi-pass? Apparently it's all about filesize predictability.

I remember that jonny once mentioned about the relationship between avg quant and compressbility. Let's say "avg quant = 4" is equivalent to "50% comp vaule". If you want your file to be 700MB, try encoding it using quant 2 first. Say it came out 1400MB in size, 1 pass at quant 4 would then be the answer for you. (Audio part is ruled out in this case). If the linear relationship is there, our encoding life would be a whole lot easier.

[cordraconis]

No, there is not such a linear relationship. If it were true, then there would only be need for 2-pass encoding: 1st to see filesize or bitrate curve, en the 2nd pass to actually encode with the exact quantisizer. (i.e. that what you are suggesting :D)
As stated before in various threads, the first-pass of divx 5 is not at maximum quality (so I'm told). Therefore, for the 2nd pass, it still makes al lot of guesses, and after the actual writing of the 2nd pass, the codec doesn't know if it's assumptions were correct.

This was also the reason for the creation of the Nth-pass feature in the DivX 5.0.3 Codec. I suggest some further reading in the threads about this feature. In theory, your encoding method should be best, but to my understanding it's not that easy.
:)

[DaveQB]

it appears i have started some good conversation,

i couldnt find anything on this topic in my searches before i posted.

as its early in the morning here and bout to leave for work, i had a wuick read of the posts but havent fully taken it in yet. But it does seem as though there isnt a hard and fast answer to this

[OvERaCiD23]

Put simply, 2-pass encoding is just to achieve maximum quality for a desired filesize (takes the amounts of bits you set, and distributes them evenly throuhgout the video to achieve constant quality throughout). 1-pass QB encoding is to achieve the same quality throughout, with no limitations on filesize. 1-pass QB @ Q2 is the highest quality achievable via DivX5, and that is a rock-solid fact.

[jonny]

The relation about average quant & comptest value is here:

http://forum.doom9.org/showthread.php?s=&threadid=42868

The last graphic show this... but my site is down :(


Anyway, for those who want to encode in quality based mode... take a look at my last DivX 5 Enc version:

http://forum.doom9.org/showthread.php?s=&threadid=22532&perpage=20&pagenumber=9

There is a totally new function called "Size prediction" :D
I think i'm the first to do something like this :D