digitalFAQ.com Forums [Archives] - Why not an easy DVD2KVCD without DVD2SVCD at all?

Page 2 of 2

Show 40 post(s) from this thread on one page

digitalFAQ.com Forums [Archives] (http://www.digitalfaq.com/archives/)

- Video Encoding and Conversion (http://www.digitalfaq.com/archives/encode/)

- - Why not an easy DVD2KVCD without DVD2SVCD at all? (http://www.digitalfaq.com/archives/encode/6877-easy-dvd2kvcd-dvd2svcd.html)

Crackhead

11-29-2003 07:37 PM

Quote:

Originally Posted by Dialhot

It seems that you didn't notice that I updated the sticky post in top of the forum with the last MA script yesterday :-)

yep, you're absolutely right :roll:

Quote:

Originally Posted by Jellygoose

@Crackhead: Please share your tests and results, I'm curious too... :roll:

will do my tests and of course :!: post the results tomorrow, since i didn't have the time to do them earlier!

Quote:

Originally Posted by Dialhot

I do a test last night and for me the result is always the same : file siez prediction of last release of D2S is completly out of target !

20 tries done, in the second one the value 72 is found as "to low" but in the 4 test the value was... 70 ! And all the tests are like this (76 is to high but three tests after I find 77...)

Finally the encoding is started because the maximum number of tests is reached. The final mpeg size was 730 Mb . 10% below the target.

I redo it currently with CQMatic and it seems that the result will be better.

I dropped D2S a long time ago and I don't think to use it again. Or I will use it for everything except video encoding part.

haha yeah you're right once again! D2S prediction is crap, but I either use azel's dvd2kvcdlite plugin if it works(will give it a try! :D ) to implement cqmatic or will do it myself and then encode with a fixed cq-value.
I'm way too lazy to do without d2s :wink:
i'm getting afraid of thinking to do every single step on my own 8O

Crackhead

11-30-2003 09:21 AM

ok finished my tests :!:
here are my settings:

Movie info:
- Dead Man / Jim Jarmusch (B&W)
- 8:03min (Chapter 1)
- Low Motion Scene / Many static objects, close-ups on faces

General Settings in D2S:
- Script: Adaptive Filtering / Optimal Script
- Audio Settings: 128kbps / Stereo
- Resolution: CVD 352x576
- Bitrate: Min 300kbps / Max 2200kbps

Encoder settings:
- RCM: CQ (Value: 80 / fixed)
- MSP: Motion estimate search
- VBV Buffer: 40
- Padding: enabled
Advanced settings:
- GOP-Structure: 1 - 25 - 2 - 1 - 25
- Detect Scene Change: Checked
- Left side: Only "Use floating point DCT" checked

ok, here the results:

File size:
Notch-Matrix: 87,6mb
Bach1-Matrix: 85,9mb
IntraShifted-Matrix: 83,1mb

Quality:

more or less equal, no noticeable difference! (at least to my and my brothers eyes :lol:)

hmm maybe the source isn't ideal for such a test, so i think i will redo it with a coloured source and high motion scenes!
any tipps or recommendations regarding the settings btw :?:

Jellygoose

11-30-2003 10:29 AM

Hi Crackhead...

I'd like to confirm your tests, but am too lazy to put the matrices in TMPGEnc manually. Could you post the part of your TMPGEnc.ini which has the matrices in it? :roll:

Crackhead

11-30-2003 11:27 AM

haha you're a lucky guy :D
since i work with d2s i don't have them in my tmpgenc.ini just in my matrices.ini of d2s. but just recently someone posted these three as part of the tmpgenc.ini :P :lol:
so here are they:

Code:

Name3="Kwag_Notch"

ReadOnly3=1

Intra3_0=8 9 12 22 26 27 29 34

Intra3_1=9 10 14 26 27 29 34 37

Intra3_2=12 14 18 27 29 34 37 38

Intra3_3=22 26 27 31 36 37 38 40

Intra3_4=26 27 29 36 39 38 40 48

Intra3_5=27 29 34 37 38 40 48 58

Intra3_6=29 34 37 38 40 48 58 69

Intra3_7=34 37 38 40 48 58 69 79

NonIntra3_0=16 18 20 22 24 26 28 30

NonIntra3_1=18 20 22 24 26 28 30 32

NonIntra3_2=20 22 24 26 28 30 32 34

NonIntra3_3=22 24 26 30 32 32 34 36

NonIntra3_4=24 26 28 32 34 34 36 38

NonIntra3_5=26 28 30 32 34 36 38 40

NonIntra3_6=28 30 32 34 36 38 42 42

NonIntra3_7=30 32 34 36 38 40 42 44

Name4="Bach_1"

ReadOnly4=0

Intra4_0=8 16 19 22 26 27 29 34

Intra4_1=16 16 22 24 27 29 34 37

Intra4_2=19 22 26 27 29 34 34 38

Intra4_3=22 22 26 27 29 34 37 40

Intra4_4=22 26 27 29 32 35 40 48

Intra4_5=26 27 29 32 35 40 48 58

Intra4_6=26 27 29 34 38 46 56 69

Intra4_7=27 29 35 38 46 56 69 83

NonIntra4_0=16 18 20 22 24 26 28 30

NonIntra4_1=18 20 22 24 26 28 30 32

NonIntra4_2=20 22 24 26 28 30 32 34

NonIntra4_3=22 24 26 30 32 32 34 36

NonIntra4_4=24 26 28 32 34 34 36 38

NonIntra4_5=26 28 30 32 34 36 38 40

NonIntra4_6=28 30 32 34 36 38 42 42

NonIntra4_7=30 32 34 36 38 40 42 44

Name5="IntraShifted by jsquare"

ReadOnly5=0

Intra5_0=8 16 19 22 26 27 29 34

Intra5_1=16 16 22 24 27 29 34 37

Intra5_2=19 22 26 27 29 34 34 38

Intra5_3=22 22 26 27 29 34 37 40

Intra5_4=22 26 27 29 32 35 40 48

Intra5_5=26 27 29 32 35 40 48 58

Intra5_6=26 27 29 34 38 46 56 69

Intra5_7=27 29 35 38 46 56 69 83

NonIntra5_0=16 19 22 26 27 29 34 37

NonIntra5_1=16 22 24 27 29 34 37 38

NonIntra5_2=22 26 27 29 34 34 38 40

NonIntra5_3=22 26 27 29 34 37 40 48

NonIntra5_4=26 27 29 32 35 40 48 58

NonIntra5_5=27 29 32 35 40 48 58 69

NonIntra5_6=27 29 34 38 46 56 69 83

NonIntra5_7=29 35 38 46 56 69 83 99

Jellygoose

11-30-2003 02:13 PM

Hi everybody...

I benchmarked the Notch Matrix against the IntraShifted Matrix and found some interesting results, which I would like to have explained and confirmed, especially by kwag...

Sample used was a part from LOTR II. Dark Scene with lots of plain surfaces, but also a face closeup and some High Action Scenes in the end... Script used was a simple Bicubic Resize script, no filters...

Compressibility:

KVCD Notch Matrix : 11.646.980 Bytes
Intra Shifted Matrix : 10.541.060 Bytes

That's a difference of 1,1 MB for a 1 minute sample :!:

Quality :

I took 3 snapshots from the same frames with VDubMod from each File and compared them.

1. DCT Blocks : Notch Matrix looks a lot better here, although far from perfect, but that's a TMPGEnc issue I'm sure. The surfaces look more flat with the notch matrix and more "unstable" with the IntraShifted...

Detail : Hard to explain. In the 3 Snapshots I took as far as Details were concerend the differences were minimal. Slight advantages to the Notch Matrix in my eyes.

High Action : Didn't see any differences here after watching the clip 10 times in a row. Might be that I'm just too tired.

Q-Curve given by Bitrate viewer.

Notch Matrix: Average: 3.32 / Peak : 4.04
IntraShifted Matrix : Average: 3.29 / Peak: 3.52

That's a surprise to me. Notice the Peak value of the Notch compared to the one of the Intra Shifted Matrix. Well it's hard to interpret these tests in my eyes. Go ahead and make tests for yourself and tell me what you guys think! :wink:

kwag	11-30-2003 03:18 PM

Quote:

Originally Posted by Jellygoose

Compressibility:

KVCD Notch Matrix : 11.646.980 Bytes
Intra Shifted Matrix : 10.541.060 Bytes

That's a difference of 1,1 MB for a 1 minute sample :!:

But results will vary, depending on the type of film.
But I'll take that one MB anytime, better than lower quality ;)

Quote:

Quality :

I took 3 snapshots from the same frames with VDubMod from each File and compared them.

1. DCT Blocks : Notch Matrix looks a lot better here, although far from perfect, but that's a TMPGEnc issue I'm sure. The surfaces look more flat with the notch matrix and more "unstable" with the IntraShifted...

That's why the "Notch" pattern was created ;)

Quote:

Detail : Hard to explain. In the 3 Snapshots I took as far as Details were concerend the differences were minimal. Slight advantages to the Notch Matrix in my eyes.

There shouldn't be any difference in high action, because of the activity. The KVCD Notch matrix will only show an advantage in low lit/dim scenes
over other matrixes. On action scenes, even though there is some differences, they are hardly visible compared to any other matrix.

Quote:

High Action : Didn't see any differences here after watching the clip 10 times in a row. Might be that I'm just too tired.

You are not tired :lol:

Quote:

Q-Curve given by Bitrate viewer.

Notch Matrix: Average: 3.32 / Peak : 4.04
IntraShifted Matrix : Average: 3.29 / Peak: 3.52

Don't trust Bitrate viewer. It's not too accurate. It even gives different average bitrate results between MPEG-1 and MPEG-2. Use Vdub to get your correct average bitrate. :)

-kwag

Crackhead

11-30-2003 03:39 PM

Quote:

Originally Posted by kwag

But results will vary, depending on the type of film.
But I'll take that one MB anytime, better than lower quality ;)

ok, but the question is if the lower cq-value of the notch matrix will result in equal or even worse quality :?:
1,1mb on 1min that's a lot! taking an 120mins movie the difference may increase up to 90mb with the same cq-value :!:

kwag	11-30-2003 03:53 PM

Quote:

Originally Posted by Crackhead

That 1MB will not be linear :!:
Not every minute will be one MB larger with KVCD's matrix.
The only way ( and correct way ) to do any comparisons ( which I have done many times ;) ) is to encode a clip with KVCD's matrix and the same clip with another matrix, and find the corresponding CQ value for each, that creates the same file size.
Then blow up your captured frame in Vdub, and look at the DCT blocks on the walls ;)

-kwag

Crackhead

12-01-2003 09:15 AM

Quote:

Originally Posted by kwag

That 1MB will not be linear :!:
Not every minute will be one MB larger with KVCD's matrix.

As I said: it might end up with such a great difference, not very likely ok, but possible :twisted: If it were linear, you would have a 132mb bigger file...lol :lol:

Quote:

Originally Posted by Kwag

The only way ( and correct way ) to do any comparisons ( which I have done many times ) is to encode a clip with KVCD's matrix and the same clip with another matrix, and find the corresponding CQ value for each, that creates the same file size.
Then blow up your captured frame in Vdub, and look at the DCT blocks on the walls

Agree on that! :D :wink:

Page 2 of 2

Show 40 post(s) from this thread on one page