All OFX spatial coordinate systems have the positive Y axis pointing up, and the positive X axis pointing right.
As stated above, images are simply some rectangle in a potentially infinite plane of pixels. However, this is an idealisation of what really goes on, as images composed of real pixels have to take into account pixel aspect ratios and proxy render scales, as such they will not be in the same space as the image plane. To deal with this, OFX has three spatial coordinate systems
The idealised image plane is always in a coordinate system of square unscaled pixels. For example a PAL D1 frame occupies (0,0) to (768,576). We call this the Canonical Coordinate System .
Many operations take place in canonical coordinates, parameter values are expressed in them while the and RoD and RoI actions report their values back in them.
The Canonical coordinate system is always referenced by double floating point values, generally via a
'Real' images, where we have to deal with addressable pixels in memory, are in a coordinate system of non-square proxy scaled integer values. So a PAL D1 image, being renderred as a half resolution proxy would be (0,0) to (360, 288), which takes into account both the pixel aspect ratio of 1.067 and a scale factor of 0.5f. We call this the Pixel Coordinate System .
The Pixel coordinate system is always referenced by integer values, generally via a OfxRectI structure. It is used when refering to operations on actual pixels, and so is how the bounds of images are described and the render window passed to the render action.
To map between the two the pixel aspect ratio and the render scale need to be known, and it is a simple case of multiplication and rounding. More specifically, given...
To map an X and Y coordinates from Pixel coordinates to Canonical coordinates, we perform the following multiplications...
X' = (X * PAR)/SX Y' = Y/(SY * FS)
To map an X and Y coordinates from Canonical coordinates to Pixel coordinates, we perform the following multiplications...
X' = (X * SX)/PAR Y' = Y * SY * FS
Note, normalised parameters and the normalised coordinate system are being deprecated in favour of spatial parameters which can handle the project rescaling without the problems of converting to/from normalised coordinates.
On most editing an compositing systems projects can be moved on resolutions, for example a project may be set up at high definition then have several versions rendered out at different sizes, say a PAL SD version, an NTSC SD version and an HD 720p version.
This causes problems with parameters that describe spatial coordinates. If they are expressed as absolute positions, the values will be incorrect as the project is moved from resolution to resolution. For example, a circle drawn at (384,288) in PAL SD canonical coordinates will be in the centre of the output. Re-render that at 2K film, it will be in the bottom left hand corner, which is probably not the correct spot.
To get around this, OFX allows parameters to be flagged as normalised , which is a resolution independant method of representing spatial coordinates. In this coordinate system, a point expressed as (0.5, 0.5) will appear in the centre of the screen, always.
To transform between normalised and canonical coordinates a simple linear equation is required. What that is requires a certain degree of explanation. It involves three two dimensional values...
As described above, the project extent is the section of the image plane that is coverred by an image that is the desired output of the project, so for a PAL SD project you get an extent of 0,0 to 768,576. As the project is always rooted at the origin, so the extent is actually a size.
Project sizes and offsets are a bit less obvious. Consider a project that is going to be output as PAL D1 imagery, the extent will be 0,0 to 768,576. However our example is a letter box 16:9 project, which leaves a strip of black at bottom and top. The size of the letter box is 768 by 432, while the bottom left of the letter box is offset from the origin by 0,77. The ASCII art below shows the details.....
(768,576) --------------------------------------- | | | BLACK | |.....................................| (768, 504) | | | | | LETTER BOXED IMAGERY | | | | | (0,72) |.....................................| | | | BLACK | | | --------------------------------------- (0,0)
So in this example...
The properties on an effect instance handle allow you to fetch these values...
So to map from normalised coordinates to canonical coordinates, you use the project size and offset...
To flag to the host that a parameter as normalised, we use the
property. Parameters that are so flagged have values set and retrieved by an effect in normalized coordinates. However a host can choose to represent them to the user in whatever space it chooses. The values that this property can take are...
kOfxParamDoubleTypeX- a size in the X dimension dimension (1D only), new for 1.2
kOfxParamDoubleTypeXAbsolute- a position in the X dimension (1D only), new for 1.2
kOfxParamDoubleTypeY- a size in the Y dimension dimension (1D only), new for 1.2
kOfxParamDoubleTypeYAbsolute- a position in the X dimension (1D only), new for 1.2
kOfxParamDoubleTypeXY- a size in the X and Y dimension (2D only), new for 1.2
kOfxParamDoubleTypeXYAbsolute- a position in the X and Y dimension (2D only), new for 1.2
kOfxParamDoubleTypeNormalisedX- normalised size with respect to the project's X dimension (1D only), deprecated for 1.2
kOfxParamDoubleTypeNormalisedXAbsolute- normalised absolute position on the X axis (1D only), deprecated for 1.2
kOfxParamDoubleTypeNormalisedY- normalised size wrt to the project's Y dimension (1D only), deprecated for 1.2
kOfxParamDoubleTypeNormalisedYAbsolute- normalised absolute position on the Y axis (1D only), deprecated for 1.2
kOfxParamDoubleTypeNormalisedXY- normalised to the project's X and Y size (2D only), deprecated for 1.2
kOfxParamDoubleTypeNormalisedXYAbsolute- normalised to the projects X and Y size, and is an absolute position on the image plane, deprecated for 1.2.
For example, we have an effect that draws a circle. It has two parameters a 1D double radius parametere and a 2D double position parameter. It would flag the radius to be
, fetch the value and scale that by the project size before we render the circle. The host should present such normalised parameters to the user in a 'sensible' range. So for a PAL project, it would be from 0..768, where the plug-in sees 0..1.
The position can be handled by the
case. In which case the plugin must scale the parameter's value by the project size and add in the project offset. This will allow the positional parameter to be moved between projects transparently.