weird artifacts on full screen display

[Francois Chartier]

Hello,

I've noticed some weird triangular artifacts on my fullscreen view in jbrout. Since that has been happening since I changed computer and made a fresh install, I thought it might be related, but since this only appears in jbrout... I thought I'd ask for other users experience...

You can see examples on issue 182 :
http://code.google.com/p/jbrout/issues/detail?id=182

These happen in full view screen:
- normal view: only if the image was rotated
- zoom view: always
and they disappear on redisplay (after I moved another window over it)

Since display is taken care of by pygtk, I guess it's not too specific to jbrout, but... any idea ?

François

[Frederic Da Vitoria]

2011/5/25 Francois Chartier <chartier...@gmail.com>

This looks a little like a video driver issue. I meant "looks" in the literal sense, because I can't figure how a video driver bug could appear only on rotated pictures :-/ What is the brand of your video board?

--
Frederic Da Vitoria
(davitof)

Membre de l'April - « promouvoir et défendre le logiciel libre » - http://www.april.org

[Frederic Da Vitoria]

2011/5/25 Frederic Da Vitoria <davi...@gmail.com>

2011/5/25 Francois Chartier <chartier...@gmail.com>

Hello,

I've noticed some weird triangular artifacts on my fullscreen view in jbrout. Since that has been happening since I changed computer and made a fresh install, I thought it might be related, but since this only appears in jbrout... I thought I'd ask for other users experience...

You can see examples on issue 182 :
http://code.google.com/p/jbrout/issues/detail?id=182

These happen in full view screen:
- normal view: only if the image was rotated
- zoom view: always
and they disappear on redisplay (after I moved another window over it)

Since display is taken care of by pygtk, I guess it's not too specific to jbrout, but... any idea ?

François

This looks a little like a video driver issue. I meant "looks" in the literal sense, because I can't figure how a video driver bug could appear only on rotated pictures :-/ What is the brand of your video board?

On second thought, the relation with rotation does not contradict a video driver issue. I don't know how it is actually implemented, but it is possible that jBrout or Python or Ubuntu actually delegated the rotation job to the video board.

[Francois Chartier]

2011/5/25 Frederic Da Vitoria <davi...@gmail.com>

This looks a little like a video driver issue. I meant "looks" in the literal sense, because I can't figure how a video driver bug could appear only on rotated pictures :-/ What is the brand of your video board?

Yes, that's what worries me slightly... it's an ATI Radeon 5450. Pretty basic board.
But I haven't noticed anything similar with other viewers, though...

 

On second thought, the relation with rotation does not contradict a video driver issue. I don't know how it is actually implemented, but it is possible that jBrout or Python or Ubuntu actually delegated the rotation job to the video board.

No, I don't think so: gnome viewer does display rotated pics on the fly, but when you apply the rotation in jbrout, the image is actually rotated (albeit losslessly). But lossless jpeg rotation may not be as innocuous as it seems :
"In particular it is possible to do 90-degree rotations and flips losslessly, if the image dimensions are a multiple of the file's block size (typically 16x16, 16x8, or 8x8 pixels for color JPEGs)."
Besides, I'm pretty sure I once read something about an effect on the borders, but that may be when this constraint on dimensions is not respected....

[Francois Chartier]

I just noticed something new, yesterday: apparently, these artefacts appear on pictures recently rotated, but not on old ones, which were probably rotated with jpegtran/exiftool, and not with exiv2... I think that might deserve a closer examination and a few tests...

François

[mana...@gmail.com]

[...] which were probably rotated with jpegtran/exiftool, and not with exiv2...

AFAIK jbrout's rotation is lossless ... and use jpegtran/exiftran

http://code.google.com/p/jbrout/source/browse/trunk/jbrout/jbrout/tools.py#530

rotation is never done with exiv2

 

--
You received this message because you are subscribed to the Google Groups "jBrout" group.
To post to this group, send email to jbr...@googlegroups.com.
To unsubscribe from this group, send email to jbrout+un...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/jbrout?hl=en.

[Francois Chartier]

No change on the way the rotation is effected in different versions of jbrout, then ? maybe something to do with my computer, then, but maybe also the way pygtk displays images. Are there some options particular to the display ?

As for the rotation, I knows it is lossless (indeed, a reverse rotation causes the artefacts to disappear). But I had read a while ago that lossless rotation of JPEG pictures could have a border effect, because of the way JPEG is encoded. It was a while ago (maybe 8 years or something), though, and I can't find that article again. I only found a reference to the dimensions which must be multiples of the block size.

2011/6/8 mana...@gmail.com <mana...@gmail.com>

[Frederic Da Vitoria]

I don't quite see how this will help us, but did you try displaying
one of the bad images on another computer?

2011/6/14, Francois Chartier <chartier...@gmail.com>:

> No change on the way the rotation is effected in different versions of
> jbrout, then ? maybe something to do with my computer, then, but maybe also
> the way pygtk displays images. Are there some options particular to the
> display ?
>
> As for the rotation, I knows it is lossless (indeed, a reverse rotation
> causes the artefacts to disappear). But I had read a while ago that lossless
> rotation of JPEG pictures could have a border effect, because of the way
> JPEG is encoded. It was a while ago (maybe 8 years or something), though,
> and I can't find that article again. I only found a reference to the
> dimensions which must be multiples of the block size.
>
>
>
> 2011/6/8 mana...@gmail.com <mana...@gmail.com>
>
>>
>> [...] which were probably rotated with jpegtran/exiftool, and not with
>>> exiv2...
>>>
>>
>> AFAIK jbrout's rotation is lossless ... and use jpegtran/exiftran
>>
>> http://code.google.com/p/jbrout/source/browse/trunk/jbrout/jbrout/tools.py#530
>> rotation is never done with exiv2

--

[Frederic Da Vitoria]

Re-reading your first mail, the remark "and they disappear on
redisplay" really means video driver issue IMO. I still can't explain
the relation to rotation, though.

What OS are you using?

2011/6/14, Frederic Da Vitoria <davi...@gmail.com>:

[Francois Chartier]

2011/6/14 Frederic Da Vitoria <davi...@gmail.com>

> I don't quite see how this will help us, but did you try displaying
> one of the bad images on another computer?

Well... yes, since that happens with all pix when zoomed in, and it didn't happen on the previous computer. changed OS version, though.

Re-reading your first mail, the remark "and they disappear on
redisplay" really means video driver issue IMO. I still can't explain
the relation to rotation, though.

What OS are you using?

I'm on Ubuntu 10.10 64 bits (used to be on 10.04 32 bits on previous computer, and am still using a 9.04 on my netbook). No problem elsewhere. Yet it could still be related to a difference in gtk version, for instance. I could try to install a different OS and see what happen...

You're probably right about the video driver, I suppose. One other thing I remarked, though, is that this happens on rotated pictures, and when zooming in but it doesn't happen if I zoom in the corners of the image, only if two borders are not visible. As long as the zoom shows the left and bottom border, for instance, there aren't any artefacts. So I wonder if we are not trying to draw outside the boundaries of the screen... these artefacts resemble indeed some things I got when playing with xlibs and messing my coordinates... maybe it could be worth a try checking the size and coordinates displayed... as for the rotation, it does change dimensions...

Thank you for your remarks, it helps progress.

Francois

[Francois Chartier]

Last news is: I tried jbrout (rpm repository version) on a Fedora 15 on my computer, with a few of my pictures and... no artefact when I zoom in or rotate. Good news is it's not material-related. I'll give a try with the svn version to see if there is any difference, and check the version number of gtk and pygtk. I could also try to upgrade my ubuntu to 11.04...

François

[matej]

Dne pátek, 17. června 2011 15:35:29 UTC+2 chartier...@gmail.com napsal(a):

I'll give a try with the svn version to see if there is any difference, and check the version number of gtk and pygtk.

For Fedora you don't have to build your own SVN checkout ... Fedora contains the latest SVN checkouts packaged https://admin.fedoraproject.org/updates/search/jbrout

[Francois Chartier]

Indeed, you're right. I didn't know the package was up to date. Not that it's difficult to use the svn version, once you have all required dependencies.

As for the effect. It seems frederic was right: no artifact on fedora... The problem must come from ubuntu's video driver for my card...

Thanks for the insights.
Francois

> --
> You received this message because you are subscribed to the Google Groups "jBrout" group.

> To view this discussion on the web visit https://groups.google.com/d/msg/jbrout/-/p5PK828_QLIJ.