Can Jonny (or anyone else) explain the basis of his CompTest prog???

[JimmyBarnes]

I know what the test is for, to try to find a suitable resolution to get a quality rip with DivX5 by sampling only 5 % of frames.

But how valid is the test? What is it's basis - I've examined the source code and it doesn't really mean much to me.

Fot the two DivX Pro 5.0 rips I've done, Jonny's CompTest v1.00 gave 49 % for Miss Congeniality and 46 % for Proof of Life. Visually the rips seem normal - high quality - even though these 1st Pass % estimates are well below the 60-75 % recommended for DivX3 rips.

I know DivX5 especially with GMC and bidirectional encoding enabled, is supposed to give good quality rips at higher resolutions for the same filesize, but ...

Anyway, your comments invited.

JB

[jonny]

CompTest works in the way described in this thread:

http://forum.doom9.org/showthread.php?s=&threadid=19927

I've done a lot of testing and seems to work well.


The last version (for DivX 5.0.1) is not fully tested.
People at DivX.com have changed the log file format (only when you enable b-frames).
In this case seems there is a "double" frame 0. I simply remove one of this "0 frame" and the results are similar to what you obtain with the old versions.

I'm going to add a compressibility test in DivX 5 Enc... this test will be more accurate...

Bye

-----
jonny

[JimmyBarnes]

Originally posted by jonny
CompTest works in the way described in this thread:

http://forum.doom9.org/showthread.php?s=&threadid=19927

I've done a lot of testing and seems to work well.


How do you calculate the maximum AVI size i.e. bitrate 6000 kbps,
DRF min and max = 2 ??

I've asked re this in this forum (how to calculate 1st Pass %) but no one has bothered to answer.

thanx
JB

[jonny]

5 - In Quick Config CLI copy:

-b21 780 -key 300 -log "C:\nnnxnnn.log" -dr 2,2,2000,10,20 -sc 50 -pq 5


min quant = max quant = 2

[JimmyBarnes]

Originally posted by jonny
5 - In Quick Config CLI copy:

-b21 780 -key 300 -log "C:\nnnxnnn.log" -dr 2,2,2000,10,20 -sc 50 -pq 5

min quant = max quant = 2

In NanDub 1.0rc2, after you had done the 1st pass Stats run and loaded it, there was a field which showed the 1st Pass AVI size in kB or MB.

Is there any similar way to read this 1st Pass AVI size from the 1st Pass VDub log file?

thanx
JB

[jonny]

This operation is actually done by GKnot (for calculating the compressibility test value), but i don't know what fields of the log file it use to do this

-----
jonny

[JimmyBarnes]

Originally posted by jonny
This operation is actually done by GKnot (for calculating the compressibility test value), but i don't know what fields of the log file it use to do this

-----
jonny

Is that info contained in the LOG file or the 1st pass AVI file?

I can work backwards to get the 1st Pass AVI size from the value your utility derives from the 13/260 sample but this is only an approximation. I like to compare the sample's 1st Pass % with that from the full run value - they are usually similar but occasionally have been significantly different.

JB

[jonny]

I'm making some testing in this direction... but using the real final avi size (not the size derived from the log file)... to see if in this way i can derive the final size (with 13/260 sample, for example)

I'll post the results (i hope in the next week)

Bye!

-----
jonny

[JimmyBarnes]

Originally posted by jonny
I'm making some testing in this direction... but using the real final avi size (not the size derived from the log file)... to see if in this way i can derive the final size (with 13/260 sample, for example)

jonny

Here's some 1-pass, 100 % quality data for 2 movies compared with your CompTest v1.00 results:

Movie Res. Scale 2-pass 1-pass 2-p,2nd CompTest
2nd pass 100 % qual as % 1-pass
=====================================================================
KTF 560x304 PC 603.4 MB 1532.3 MB 39.6 % 50.9 %
POL 560x240 TV 598.9 MB 1465.4 MB 41.2 % 45.9 %

In all cases I used:
* DivX Pro v5.0
* GMC and bidirectional encoding in 2-pass runs
* GMC only in 1-pass runs
* The frameserver AVI directly rather than via an AVS file
* Resizing and Full Processing Mode in VDub 1.4.10.
* Precise bicubic resizing in VDub

Just for interest, for KTF I did a NanDub 1.0rc2 stats run at the KTF resolution above with precise bicubic resizing. When fed to GKnot 0.23b, this yielded a 1st Pass AVI size of 1513.4 MB.

Even though 2-pass, 2nd pass sizes when compared to 1-pass 100 % quality sizes were low (around 40 %), video quality was good/very good (I am experienced with DivX3.11a ripping). The blockyness (not macroblocks) seen on large uniform areas like walls or sky which spoils otherwise good DivX3.11a rips was minimised with DivX5.

JB