Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
22nd September 2009, 18:42 | #581 | Link | |
Moderator
Join Date: Nov 2001
Location: Netherlands
Posts: 6,364
|
Quote:
|
|
8th October 2009, 23:42 | #583 | Link |
契約者
Join Date: Jun 2008
Posts: 1,576
|
Hi. I'd like to ask something about colormatrix...
Here is info from DGIndex about my DVD. Here is a screenshot, taken in VirtualDub using avisynth script with only one line in it: Mpeg2source("VTS_01_1.d2v") Here is a screenshot of the same frame, but taken with following script: Mpeg2source("VTS_01_1.d2v",info=3).Colormatrix(hints=true,interlaced=true) It looks like screenshots are the same, but here is what I see after overlaying them using "difference" and boosting levels. I expected to see no difference at all or to see much more difference... So all is fine here, this is how it should be? Am I doing something wrong? |
9th October 2009, 01:32 | #584 | Link |
契約者
Join Date: Jun 2008
Posts: 1,576
|
After reading many nice posts by manono in this thread, I think maybe the difference on my screens appears because of clamping some levels after applying colormatrix. Indeed such things shouldn't be done 'behind the back'.
Last edited by Keiyakusha; 9th October 2009 at 01:46. |
30th October 2009, 02:11 | #586 | Link |
Kid for Today
Join Date: Aug 2004
Posts: 3,477
|
XhmikosR has been kind enough to compile an ICL11/SSE3 version of ColorMatrix, on the aforementioned call(on 4 threads w/ a 3.2Ghz Q6600) I get a 3% speed improvement(going from 593 to 609 fps): http://www.mediafire.com/?n2gumty1zyn
|
30th October 2009, 02:25 | #587 | Link | |
契約者
Join Date: Jun 2008
Posts: 1,576
|
Quote:
|
|
30th October 2009, 13:49 | #588 | Link | |
Moderator
Join Date: Nov 2001
Location: Netherlands
Posts: 6,364
|
Quote:
|
|
30th October 2009, 13:51 | #589 | Link |
Registered User
Join Date: Mar 2007
Location: Hellas (Greece)
Posts: 543
|
So you need the ICL project files? Because that's the only thing that's different and that I changed the File Info to distinguish it from the normal build.
Anyway, here is the package with the source included. Last edited by XhmikosR; 3rd November 2009 at 22:25. |
12th December 2009, 16:32 | #590 | Link |
Registered User
Join Date: Jan 2009
Posts: 251
|
Feature request: 64 bit version
Thanks for your great tool. I use it all the time.
I was wondering if it would be feasible to prepare a 64-bit version. I have been looking at squid_80's 64 bit version of AVISynth and related 64-bit plugins that he has prepared, and he has everything I use in AVISynth ready in 64-bit except ColorMatrix. I have no idea what that would imply, but if it is something feasible it would be great! Thanks again. |
25th December 2009, 22:44 | #591 | Link |
Kid for Today
Join Date: Aug 2004
Posts: 3,477
|
hi tritical, I was wondering if you had any idea about this: http://forum.doom9.org/showpost.php?...postcount=1751
problem of chroma alignment possibly? any idea how to get the same exact output as ffdshow please? |
26th December 2009, 14:40 | #592 | Link |
Registered User
Join Date: Sep 2004
Posts: 1,295
|
When you use t3dlut, the conversion between BT.601 and BT.709 coefficients is performed like this:
1) up-sample chroma from 4:2:0 or 4:2:2 to 4:4:4 2) convert BT.601 -> BT.709 3) sub-sample chroma from 4:4:4 to 4:2:0 or 4:2:2 I'm not sure about Colormatrix, but I think it will work similarly... When you use ffdshow it's a different thing, because you are not converting YV12->YV12 or YUY2->YUY2, you are converting YV12->RGB32, and in this kind of conversion, you don't need to downsample the chroma at the end. Summing up, you only could be able to get the same output as ffdshow if you use it like ffdshow. With t3dlut, you should set it for YV12->RGB32, and not for YV12->YV12 as you are performing in your examples. With Colormatrix, that's not possible, because it only performs conversions between YV12/YUY2->YV12/YUY2. Last edited by yesgrey; 26th December 2009 at 15:08. |
26th December 2009, 14:46 | #593 | Link |
Kid for Today
Join Date: Aug 2004
Posts: 3,477
|
ok, but we had the same problem when I wanted to get the same output from your LUT as I was getting from the PS script...and at some point, we did! you told me there was some hiccup in the LUT code and made a bugfix
I'll try to fix the 601>709 LUT conversion .ini file, and hopefully I'll have more details about what's "broken" to give to tritical. |
26th December 2009, 15:08 | #594 | Link | |
Registered User
Join Date: Sep 2004
Posts: 1,295
|
Quote:
From what I understand, the difference between Colormatrix and t3dlut is that the later let's the user define which chroma placement to consider, while the former always assume it internally. I believe the problem is due to the last step of chroma sub-sampling, so I don't think it could be fixed, because there is nothing to fix... |
|
28th December 2009, 18:34 | #597 | Link | |
Kid for Today
Join Date: Aug 2004
Posts: 3,477
|
Quote:
but if you compare to these, ffdshow still looks slightly better IMHO: http://forum.doom9.org/showpost.php?...postcount=1751 still would be pretty cool if tritical could allow to change the chroma placement in ColorMatrix() Last edited by leeperry; 28th December 2009 at 18:51. |
|
28th December 2009, 20:35 | #598 | Link |
Registered User
Join Date: Dec 2003
Location: MO, US
Posts: 999
|
To match ffdshow you probably need to set itypeU=1 in t3dlut. Since by default t3dlut uses cubic interpolation for 4:2:2->4:4:4, which utilizes 4 pixels, so the end points of the kernel will still be spanning those color blocks. Anyways, I think basing what looks best on that one image is not a good idea. It contains big blocks of color with hard transitions and mpeg1 chroma placement... so of course duplication and mpeg1 chroma placement assumption in t3dlut/yv12toyuy2 will look best on it. However, I could easily generate an image where that combination will look much worse than cubic interpolation with mpeg2 chroma placement assumed. Anyways, with respect to colormatrix there is nothing to change in regards to chroma alignment since it does not upsample. It converts directly YUV->YUV, to get the UV values for computing the new Y samples it uses nearest neighbor. That is, in YV12 each set of 4 luma pixels ((x,y)(x+1,y)(x,y+1)(x+1,y+1)) uses the U and V pixels at (x>>1,y>>1) in the U and V plane respectively.
|
Tags |
colormatrix |
Thread Tools | Search this Thread |
Display Modes | |
|
|