Filter comparisons

sparky · Joined: 10 Jun 2007 Posts: 21

JJ
Updated .png screenshots, is this what your after.

JJ · Posted: Fri Nov 20, 2009 10:05 pm Post subject:

Sorry but these shots are useless in comparison. You need to use exactly same resolution as DVD, 720x480 or 720x576.
Now your software player is resizing picture to bigger resolution and it changes picture.

mike22490 · Joined: 27 Sep 2009 Posts: 41

Honestly, if you want the most quality you can get use AvstoDVD

Source: http://www.microsoft.com/windows/windowsmedia/musicandvideo/hdvideo/contentshowcase.aspx

Step Into Liquid 1080p

Convertxtodvd 4.13 2 Pass HQ (under 80 Minutes) Lanzcos Conversion Completed In 2:19

AVStoDVD 2.24 FFmpeg+Aften Audio Encoder, Directshowsource filter, Spline36 (Custom) Filter, Hcenc 2 Pass
Conversion with 0.2 sharpning applied Completed In 4:38

Picture 1 Is Convertxtodvd

Picture 2 Is Avstodvd

Let your eyes decide...

P.S. I challenge any one to side-to-side video comparisons.

You pick the source... and use convertxtodvd for your encode, and ill use avstodvd....

I bet avs will come out on top every time...

hansK · Posted: Fri Nov 27, 2009 11:33 am Post subject:

mike22490 · Joined: 27 Sep 2009 Posts: 41

Here's a comparison with the same setting's used...

Spline Filter (Converxtodvd Spline) AVStodvd (Spline36)
2 Pass Encode
And No Extra Sharpening Applied

Let your eyes be the judge....

All screen shots start with convertxtodvd then AVStoDVD

So as you can see not only does Avstodvd have higher quality levels... but the color maintained is also higher then convertxtodvd....

Cant wait to see what arguments I have to defend next...

BTW... I compared convertxtodvd's Lanzcos against avstodvd's spline36 and avstodvd was superior in quality every time...

mike22490 · Joined: 27 Sep 2009 Posts: 41

Here's another 2 with same settings...

Should I even point out which one is covertxtodvd?

A joke if you ask me...

One's Free and ones 50 bucks...

admin · Site Admin Joined: 18 Jul 2006 Posts: 16777215

great contribution, you did an amazing job with all the screenshots,
you are right, your solution is better. maybe it is the sharpening part which make the difference. I m not the expert on the topic

mike22490 · Joined: 27 Sep 2009 Posts: 41

thank you...

By the way no sharpning was applied in any of the screenshots...

Personally I usaully do add 0.2 sharpning on all my conversions with avstodvd, it gives it a very very nice look...

But as you can see above, the sharpening is not even needed..

Please note I'm not trying to bash convertxtodvd...

Only Improve it...

P.S. Out of all the filters ive found Spline36 to produce the clearest, sharpest image... (better then Lanzcos for both upsizing and downsizing)

If convertxtodvd could add this filter I'm sure there would be a great quality increase...

The mathematical equation can be found online...

ckhouston · Joined: 08 Aug 2007 Posts: 2008

mike22490:

Until you slow down, pay attentention to what you and other posters are told, and develop a systematic approach to comparing CX2D with avstodvd, I recommend that people ignore your seemingly impressive screenshots.

It has been said many times before that screenshots of conversions do not provide valid comparisons because they cannot ususlly be done at the same frame. When comparing results with the same program, the same frame can be compared, but only if frame captures from a good video editor are made for the same frame, it can't be done with a software player. When comparing two programs, one may make captures at the same frame, but each program converts differently, so captures of the same frame may still not be valid because the programs may be applying compression for different frame types (I,P or B) at that point. Quality of the different frame types varies considerably as you can see in a video editor by advancing frame-by-frame. So, when comparing different programs, it is usually best to compare the same frame type even if the time of the frames is slightly different.

And I have said before that I have three PCs with LCD monitors and all of them are darker at the top of the screen than the bottom. So comparisons of images arranged vertically are difficult to evaluate.

I downloaded two of your images, cropped the right half side of both and arranged them side by side in the attached image -- Cx2D is on the left followed by avstoDVD. That solves the difference in brightness for me.

The avstodvd image is definitely better, BUT, compare distances from the left side to the girl and that from the right side to the red cone and it is evident they are not the same frame -- the camera is panning. So that comparison is useless IMO.

The preponderance of better quality for avstodvd for all your comparisons does indicate thatr avstodvd may be better in spite of problems I mentioned. BUT, ONCE AGAIN, there is another important factor that you have ignored. We have no idea what settings you used for CX2D or for avstodvd other than the resampling filter. There are many settings besides resampling filters that can influence quality considerably for both programs. In fact, I can make CX2D look better than almost any program and vice versa by manipulating settings intentionally.

So I again recommend that readers ignore your seemingly impressive image comparisons until you do develop a systematic approach and give us enough info to duplicate and evaluate your comparisons. And I recommend, as edmantay already did in another thread, that you start a new independent thread that can be devoted to a logical investigation of the two programs.

mike22490 · Joined: 27 Sep 2009 Posts: 41

Regarless that the frame was off by 1 or 2 in the picture you posted I have many other screenshots to back it up...

And settings used is simple...

Covertxtodvd: Spline 2 Pass With the Latest Version...

AVSTODVD:

ckhouston · Joined: 08 Aug 2007 Posts: 2008

Ok I'm going to download avstodvd and do a comparison of my own.

Two questions:

1. Was your source DVD compliant?

2. What were your settings with CX2D? How about pasting your CX2D log for that conversion so we can see what you did?

mike22490 · Joined: 27 Sep 2009 Posts: 41

http://www.microsoft.com/windows/windowsmedia/musicandvideo/hdvideo/contentshowcase.aspx

Source is 1080p Step Into Liquid..

I have no logs since I uninstalled cx2dvd

It's not like convertx has a tone of options... try both lanzcos and spline...

Avs will come on top....

(make sure to add the spline36 filter to avstodvd)

admin · Site Admin Joined: 18 Jul 2006 Posts: 16777215

just for your information, we are using that reference file ( because it is a free movie and not too big to download )

http://download2.vso-software.fr/other/big_buck_bunny.avi

so we could share also the same file to do test.

admin · Site Admin Joined: 18 Jul 2006 Posts: 16777215

also , could I suggest a still photo resizing application we have to appreciate live resize impact ?

http://download2.vso-software.fr/vso_photo_gallery_setup.exe

first go to main menu options / File management, and turn on the option on the bottom 'debug mode'

then navigate to the folder with the screenshots, double click to open the full view, and in the bottom you will see the list of possible filters ( vso lanczos etc.. )
the filter will be applied in live, during the resize and on load.

ckhouston · Joined: 08 Aug 2007 Posts: 2008

mike22490 · Joined: 27 Sep 2009 Posts: 41

I used "short play" under 80 minutes encoding option for convertx

For the screenshots I used VLC player and tryed to match them up as best a could...

dictator · VSO Team Member Joined: 11 May 2007 Posts: 54

Hello Mike22490,
hello all

First, I need to say thank you for your investigation which is very interesting.

I haven't had the time yet to investigate extensively about this, however IMHO it's not related to the filters.

For our filter, it's correct to say that they currently use a smaller window that a Spline36 one (19 pixels instead of 36), and maybe ( ? ) a smaller precision to do the computation (we use 14 bits, spline36 ? ).
However, What I can certify is that our filters are color accurate, which mean they don't introduce a tint difference, which is often the case in such filters due to rounding issues.

So, difference in color saturation can't be explained by the usage of our filters. If it's not from the filters, it can come from any other part, being :

decoding:
your software do use DirectShow source. This mean it uses the official decoder embedded with Windows. By default, convertX use internal decoding engine, which allow to be 100% independant and work even if you don't have anything installed on your computer to decode your video.

So, in your case we don't use the same decoder.

image processing:
the fact that it uses DirectShow source mean your decoder can be set with many image processing, such as a sharpening, color saturation, etc etc. The only image processing ConvertX use (for now) is deinterleaving, when necessary.

so we also don't use the same image processing

encoding:
We use the FFMpeg encoder, tuned for motion detection and bitrate control to obtain the best balance between speed and quality, and to ensure the best compliance with the DVD specs.
Is it the same encoder as the one used by your software ? I don't know.
For ConvertX, there is many settings possible to achieve higher quality as well as better psycho-perception of the image, however all of these settings may have a dramatic impact on encoding speed. So we made a choice achieving the best balance.

So, IMHO again the more probable explanation for such a difference is related to amount of image processing provided by the DirectShow layer, which in your case has probably provided sharper and more colorfull images.
I am almost sure that if ConvertX is fed with the same image, our result will be as good as yours.

The immediate question leading to this is "ok, so why don't convertX uses DirectShow as image provider ?". The reason is that we never know for sure what the DirectShow output is going to provide. In some case, it can be pristine quality (especially when it's hardware decoders embedded in good video card that give you the data), in other case it can be really poor and blocky, or output nothing at all if the codec is not installed in the window's system.
So that's why there is no setting asking to use DirectShow output in ConvertX, and directShow is used by ConvertX only if there is no other way to decode.

We are really concerned by the output quality in ConvertX, however we also know that even insane quality settings may improve the quality only by a few percent, whereas the encoding process may take up to 20 time longer. This is not something that everyone is ready to accept.

And about this specific issue, it can be solved by adding some nice image postprocessing such as sharpening and color enhancement, rather than providing yet another resize filters which is imho not the cause of the problem here.

Bags · Posted: Sun Nov 29, 2009 4:28 pm Post subject:

Hi Jacques ...

Well that is enough for me ... a well presented defence strategy there!

I'm personally un-convinced by using too many processing effects (especially sharpening) This can often introduce deterioration of a different kind (like contrast, colour, etc) to displayed images (obvious on my Pioneer Plasma screen)

I have always found CXDVD to do the best job (reliably) with relation to time and source quality. It will probably also convert more formats/standards than any other program that I know of.
Sure, programmes like TMPGenc may be able to improve overall quality by a few percent, but at the expense of something else.

Regards ....

raymondtrudeau · Joined: 19 Nov 2008 Posts: 12

Bags · Posted: Sun Nov 29, 2009 8:14 pm Post subject:

Why ask that here? Is this the AVS Forum?
There is online help link (and guides) on their website.