- From: Michael Dolan <mdolan@newtbt.com>
- Date: Mon, 18 Apr 2016 18:31:12 -0700
- To: "'TTWG'" <public-tt@w3.org>
- Message-ID: <03e701d199db$29cddd60$7d699820$@newtbt.com>
Sorry, Glenn. I was not appreciating the new semantics in TTML2. My head was still around TTML1 semantics, not TTML2 and the new semantics around div and bpd. Now that I have read it :) …
Yes, it is very safe to assume that the DECE validator does not check the details from your example.
This new behavior is outside the original high level requirements for image subtitles. The original requirements were about enabling pixel-level authoring control of glyphs specifically including the management of the extent of the enclosing area of rendered glyphs (especially horizontal extent). Although originally conceived to mainly address non-Latin fonts, I am told that there is growing interest in the use of image subtitles for even Latin languages in cinema applications. Maybe the requirements have more recently expanded? Maybe it would be good to make blocks/lines of images scroll like text? (I’m just guessing here).
I would recommend reaching out to potential users and ask them how they’d like it to behave.
Mike
From: Glenn Adams [mailto:glenn@skynav.com]
Sent: Monday, April 18, 2016 4:53 PM
To: Michael Dolan <mdolan@newtbt.com>
Cc: TTWG <public-tt@w3.org>
Subject: Re: [ttml2] Does tts:backgroundImage affect size of content area?
On Mon, Apr 18, 2016 at 4:51 PM, Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> > wrote:
>> BTW, what do authors do to satisfy the DECE CFF validator? i.e., how to they ensure the content size matches?
The “package” in this case is an ISOBMFF file coded according to ISO 14496-30 and DECE CFF. The TTML document and images are in a sample using smpte:backgroundImage=”URN:…” to reference them within the sample.
DECE/IMSC1 made a simplifying constraint on SMPTE-TT where a <region> has exactly one <div> that has exactly one image file reference. smpte:backgroundImageHorizontal and smpte:backgroundImageVertical are forbidden. So, the TTML region extent is compared to the extent of the image file that is referenced within the one <div>. There are not multiple <div>’s per region, and thus not multiple image files per region.
ok, but that test does not test whether the content element (div) has a content rectangle that matches the size of its background image;
that test, only tests if the background image of some content element selected into the region will fit or match the geometry of the region;
as I have pointed out, the content rectangle of a div without any content is empty (by XSL-FO/CSS rules). So the background image will not be displayed unless, and only unless you do one of the following:
1. add content to the div, e.g., significant whitespace, that provides adequate extent in the bpd dimension to encompass the image's bpd;
2. specify bpd (or height in horizontal WM) on the content element (div) that matches or exceeds the background image's bpd;
3. deviate from XSL-FO/CSS rules, e.g., by requiring the background image size to affect the content rectangle size of the content element (div) on which the background image is specified;
Option 1 won't work in IMSC, since you can't mix text with images. Option 2 won't work with either ST2052 or IMSC, since there is no bpd (height) property to specify. Option 3 seems the only option and is supported by the language I cite from ST2052 about making the minimum content rectangle of div match the background image size.
Option 1 might look like:
<region xml:id='r1' tts:extents='100px 100px'/>
...
<div region='r1' smpte:backgroundImage='100x100.png'>
<p tts:fontSize='100px' tts:lineHeight='100px'> </p>
</div>
Here, we add content to the div, a single NBSP character, with height 100px, to ensure the bpd of the content rectangle is (at least) 100px. The ipd of the content rectangle of the div gets the ipd from the region (100px), so we have matched the size of the content rectangle of the div to the background image size.
I'm sure that the DECE CFF validator doesn't test this.
Option 2 might look like:
<div region='r1' smpte:backgroundImage='100x100.png' tts:bpd='100px'/>
of course there is no tts:bpd in ST2052 or IMSC1.
In TTML2, one could have either:
Option 1'
<div region='r1'>
<image src='100x100.png'/>
</div>
This works by having a block level content image as the content of div, which does use the intrinsic size of the image to generate an block area child of the div's block area.
or
Option 2'
<div region='r1' tts:backgroundImage='100x100.png' tts:bpd='100px'/>
This works since we can explicitly style the div's bpd (height here) in TTML2.
Regards,
Mike
From: Glenn Adams [mailto:glenn@skynav.com <mailto:glenn@skynav.com> ]
Sent: Monday, April 18, 2016 2:05 PM
To: Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> >
Cc: TTWG <public-tt@w3.org <mailto:public-tt@w3.org> >
Subject: Re: [ttml2] Does tts:backgroundImage affect size of content area?
On Mon, Apr 18, 2016 at 2:51 PM, Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> > wrote:
I see. I did not note the zero height div. But there is no text writing in the IMSC1 image profile (text is forbidden).
Exactly. So in IMSC1, even if you wanted to match the content size to the background image size, you don't have any way to do that explicitly, other than by relying on the semantics of ST2052-1:2013 with respect to making the minimum width and height be at least as large as the background image size (which is at variance with CSS).
Required extent matching was the intent in ST 2052-1 regardless of how softly it was stated. And, FYI the DECE CFF validator would reject images that do not match.
If it is required to match, then I wonder why the language
The min-height and min-width of the area associated with the image shall be set to the intrinsic height and width of the image source.
This effectively makes the content size matching requirement rather pointless.
BTW, what do authors do to satisfy the DECE CFF validator? i.e., how to they ensure the content size matches?
I was only trying to point out that a 3rd option is to forbid the mismatch.
I don't think we need to do this in TTML2, since there are well defined XSL-FO/CSS semantics for how to handle this case. In any case, I don't see it being forbidden by ST2052-1:2013 or IMSC1, regardless of whether the DECE validator would fail that content. I don't even see how the DECE validator can check this constraint for that matter.
If that still doesn’t work, that’s fine. Of the other options you have proposed, I don’t have a preference.
Regards,
Mike
From: Glenn Adams [mailto: <mailto:glenn@skynav.com> glenn@skynav.com]
Sent: Monday, April 18, 2016 1:23 PM
To: Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> >
Cc: TTWG <public-tt@w3.org <mailto:public-tt@w3.org> >
Subject: Re: [ttml2] Does tts:backgroundImage affect size of content area?
On Mon, Apr 18, 2016 at 1:53 PM, Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> > wrote:
Perhaps I am missing something subtle. I thought you were trying to prescribe decoder behavior when the target content element size does not match the referenced image size. This condition is an authoring error in SMPTE-TT, so the behavior is undefined (although arguably the decoder could reject the package).
The subtlety is that a content element, e.g., div, has a size of zero in the block progression dimension (height in horizontal writing modes). So if that div has a background image, it will not be displayed (in XSL-FO and CSS).
However, ST2052-1:2013 does something different, when it states (in §5.5.6):
The min-height and min-width of the area associated with the image shall be set to the intrinsic height and width of the image source.
The example I included in the issue [1] shows that CSS does not use this behavior.
[1] https://github.com/w3c/ttml2/issues/157
So, as we introduce tts:backgroundImage to TTML2, it is natural that we consider whether to adopt XSL-FO/CSS behavior (my preference) or SMPTE-TT/IMSC1 behavior.
Also, the above cited section in ST2052-1 also goes on to say:
authors should ensure that the div will be sized to match the given pre-rendering
But it doesn't mandate this or make it invalid to fail to do this. Further, it begs the question of how one would ensure the div will be adequately sized. Since ST2052-1 (nor TTML1) defines a way to explicitly size a content element (like div), one wonders how one can satisfy this condition [that the div will be sized to match].
If TTML2 wishes to allow this situation and thus define decoder behavior for it, OK.
I don't see where ST2052-1 or IMSC1 prohibits "this situation" as you have described it. Just the opposite, I don't see how one can ensure the size of the content matches or exceeds the background image size. For ST2052-1, you could try to include invisible text content (by using tts:visibility of 'hidden'), but it will be difficult to get that text size to match the background image size. In IMSC1 you can't even do this, since you can't include text at all.
Then I guess I don’t have a position. Alternatively TTML2 could also declare that condition to be an error and therefore not address it in the first place, which is what I was trying to suggest.
Mike
From: Glenn Adams [mailto: <mailto:glenn@skynav.com> glenn@skynav.com]
Sent: Monday, April 18, 2016 12:21 PM
To: Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> >
Cc: TTWG <public-tt@w3.org <mailto:public-tt@w3.org> >
Subject: Re: [ttml2] Does tts:backgroundImage affect size of content area?
I'm afraid I'm still not following you. My comment is not related to validity, but the semantics of background image impact on sizing the content element that refers to it.
SMPTE-TT says that the minimum size of the area generated by the content element is expanded to fit the size of the background image. This is not the behavior in CSS or XSL-FO, so it presents an issue for how to deal with background image in TTML2 and how to translate SMPTE-TT or IMSC1 to TTML2.
My current thinking is that tts:backgroundImage in TTML2 will not have the behavior of SMPTE-TT or IMSC1, i.e., it will follow the XSL-FO and CSS behavior. So if background image is used in TTML2, the author will need to ensure the size of the generated area is sufficiently large, e.g., by including sufficient content or by using tts:ipd or tts:bpd.
For the purpose of translating SMPTE-TT or IMSC1 to TTML2, a couple options exist:
* translate to <image> element rather than tts:backgroundCheck
* or use tts:backgroundCheck but also specify tts:{ipd,bpd}
I don't see any issue here with respect to validity.
On Mon, Apr 18, 2016 at 11:59 AM, Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> > wrote:
The IMSC1 instance document alone can be valid IMSC1. But in the context of it with the collection of (e.g. PNG) image file properties, the “package” is not valid. Just because we don’t have a formalization of the “package” doesn’t mean that a decoder should adopt some compensation behavior when faced with such a “package”.
Mike
From: Glenn Adams [mailto:glenn@skynav.com <mailto:glenn@skynav.com> ]
Sent: Monday, April 18, 2016 9:26 AM
To: Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> >
Cc: TTWG <public-tt@w3.org <mailto:public-tt@w3.org> >
Subject: Re: [ttml2] Does tts:backgroundImage affect size of content area?
Could you please explain? What is invalid?
On Mon, Apr 18, 2016 at 8:46 AM, Michael Dolan <mdolan@newtbt.com <mailto:mdolan@newtbt.com> > wrote:
This condition is an invalid document and thus decoder behavior is undefined.
Mike
-----Original Message-----
From: Glenn Adams via GitHub [mailto:sysbot+gh@w3.org <mailto:sysbot%2Bgh@w3.org> ]
Sent: Saturday, April 16, 2016 1:47 PM
To: public-tt@w3.org <mailto:public-tt@w3.org>
Subject: [ttml2] Does tts:backgroundImage affect size of content area?
skynavga has just created a new issue for
https://github.com/w3c/ttml2:
== Does tts:backgroundImage affect size of content area? == According to SMTPE ST2052-1:2013 §5.5.6:
> The min-height and min-width of the area associated with the image
shall be set to the intrinsic height and width of the image source.
and
> The presented image shall not be scaled and the XSL background-color
trait shall be visible for any background areas of the ``<div>``
outside the image, therefore, authors should ensure that the div will
be sized to match the given pre-rendering.
The effects of these two requirements are that if an image is larger
than the nominal minimum content size of the generated area, then that
minimal content size is to be increased to encompass the intrinsic
image size as required. While if the nominal minimum content size is
already larger than the intrinsic image size, the background color of
the generated area (if not transparent) will be visible outsize the
image, i.e., the image is not scaled in either dimension to fill the
nominal minimum content size in such a case.
IMSC1, which makes use of the ``smpte:backgroundImage`` extension,
does not specify any details about sizing, other than to say that the
intrinsic image size must correspond to the size of the region into
which its associated generated area(s) are flowed. As such, one may
presume that IMSC1 intends the above language from ST2052 to apply
(Pierre?).
The problem here is that this behavior is at variance with both XSL-FO
and CSS, where the minimum content size of generated areas associated
with a background image **is not** increased to the intrinsic image
size. In other words, in XSL-FO or CSS, if there is no content in the
element, then no area is generated, or if one is generated, it has
zero for one or both content area dimensions. On the other hand, if
there is content in the element, but its nominal formatted size is
smaller than the intrinsic size of the background image, then some of
the image will be clipped. I am attaching a test file that
demonstrates the CSS behavior below.
Now, the problem here in TTML2 is how to handle this discrepancy
w.r.t. XSL-FO and CSS introduced by SMPTE-TT and apparently adopted by
IMSC1. One way to handle it would be to require that ``tts:ipd`` and
``tts:bpd`` be used with TTML2 if one wants such behavior with a
background image (in a like manner to how one is forced to specify
width and height properties in XSL-FO or CSS). Or we could add a
``tts:{ipdMinimum,bpdMinimum}`` pair in TTML2 to specify instead. Or
we could generalize the values of ``tts:{ipd,bpd}`` to allow
specifying a (minimum, optimum, maximum) tuple. Or we could allow the
minimum values of these properties to be specified with a keyword, say
``backgroundImageSize``.
In any case, we have an issue here, so I have added an editorial note
to the section on
(tts:backgroundImage)[https://rawgit.com/w3c/ttml2/master/spec/ttml2.html#style-attribute-backgroundImage]
in the TTML2 ED.
Following is the CSS example file:
[test.html.zip](https://github.com/w3c/ttml2/files/222397/test.html.zip)
Please view or discuss this issue at
https://github.com/w3c/ttml2/issues/157 using your GitHub account
Received on Tuesday, 19 April 2016 01:31:47 UTC