Re: [filter-effects] resolution dependent filter primitives

On Sep 24, 2013, at 6:01 PM, Stephen Chenney <schenney@chromium.org> wrote:

> 
> 
> 
> On Wed, Aug 28, 2013 at 1:02 AM, Paul LeBeau <paul.lebeau@gmail.com> wrote:
> On 28 August 2013 16:38, Dirk Schulze <dschulze@adobe.com> wrote:
> 
> > Is the blurring following the rotation or does it stay in the horizontal position?
> 
> The latter.  The blur is applied to the transformed object, so even if stdDevation.Y is 0, you still get blurring in the "Y" direction.
> 
> Paul
> 
> Surely this is because they apply the blur to the frame-buffer target in hardware (not to a render target that is subsequently applied as texture to a transformed quad).

I do not speculate about the reason for that. It is not following the current specification though. However, in certain situations you want this kind of behavior. Imagine you have a rotating planet with a shadow. Then you may not want the shadow to rotate as well. Similar examples are possible with blur, where the object rotates but you want the blur radius to be horizontal no matter what.

I think a shadow operation in screen coordinates can be desirable in some cases and is a great addition for level 2.

Greetings,
Dirk

> 
> Stephen.
>  
> On 28 August 2013 16:38, Dirk Schulze <dschulze@adobe.com> wrote:
> 
> On Aug 28, 2013, at 6:11 AM, Paul LeBeau <paul.lebeau@gmail.com> wrote:
> 
> >
> > On 27 August 2013 20:42, Dirk Schulze <dschulze@adobe.com> wrote:
> > Again, it is extremely hard to determine the right scale level for the horizontal and vertical axis...
> >
> > As someone who is implementing filter support in my renderer at the moment, I can concur with this :)
> >
> > It would be helpful if the spec gave some guidance on best practice for determining an appropriate size for the filter bitmap.  By "filter bitmap" I mean the intermediate bitmap used for SourceGraphic etc.  For axis aligned content it is easy, but if the element has a transform, it is not so straightforward.
> 
> I actually planned to discuss that in a different threat. I agree with you on that and invite implementers to give feedback on how they estimate the affected area in screen coordinates. I will do so for WebKit shortly.
> 
> >   In fact I notice that IE doesn't even try. It performs all filters on the transformed content and doesn't use user-coordinate-space.
> 
> I never really experimented with filters in IE so far. I wonder how that works with gaussian blur and a horizontal blur radius on a rotated object. Is the blurring following the rotation or does it stay in the horizontal position?
> 
> Greetings,
> Dirk
> 
> >
> > Paul
> >
> 
> 
>

Received on Tuesday, 24 September 2013 17:24:45 UTC