[fusion] Defining pixel formats
Hi all, ever since Jang Tongary suggested to use boost::fusion to define pixel types I wondered what he meant by that. I never had the opportunity to get myself familiar with fusion so I started learning it and I'm really amazed how easy the library is to use. I think I'm at a point where I would like to reach to the community for some advice. Let's say I wanna create a homogeneous three channel pixel type, like RGB. For that I would use a fusion map since I want to access channel by name. I also would like to be able to swap channels around to create BGR. How would I do that? namespace boost { namespace gil_3 { struct red_t; struct green_t; struct blue_t; } } // namespace boost::gil_3 void foo() { typedef fusion::map< fusion::pair< gil_3::red_t , boost::uint8_t > , fusion::pair< gil_3::green_t, boost::uint8_t > , fusion::pair< gil_3::blue_t , boost::uint8_t > > rgb8_pixel_t; // not working typedef fusion::reverse_view< rgb8_pixel_t >::type bgr8_pixel_t; rgb8_pixel_t a( 10, 20, 30 ); bgr8_pixel_t b( 30, 20, 10 ); boost::uint8_t red_a = fusion::at_key< gil_3::red_t >( a ); boost::uint8_t red_b = fusion::at_key< gil_3::red_t >( b ); } Thanks! Christian
2013/3/17 Christian Henning
Hi all,
ever since Jang Tongary suggested to use boost::fusion to define pixel types I wondered what he meant by that.
I believe you're mentioning me here. ;-)
I never had the opportunity to get myself familiar with fusion so I started learning it and I'm really amazed how easy the library is to use. I think I'm at a point where I would like to reach to the community for some advice.
The very idea is to use Fusion to manipulate the pixel, not defining the pixel, though you could do it. I have one file "adapted_pixel.hpp" that adapts GIL pixel (concept) to Fusion Associative Sequence. Though I want to share here but there's another dependency on my other header.
Let's say I wanna create a homogeneous three channel pixel type, like RGB. For that I would use a fusion map since I want to access channel by name. I also would like to be able to swap channels around to create BGR. How would I do that?
namespace boost { namespace gil_3 {
struct red_t; struct green_t; struct blue_t;
} } // namespace boost::gil_3
void foo() { typedef fusion::map< fusion::pair< gil_3::red_t , boost::uint8_t > , fusion::pair< gil_3::green_t, boost::uint8_t > , fusion::pair< gil_3::blue_t , boost::uint8_t > > rgb8_pixel_t;
// not working typedef fusion::reverse_view< rgb8_pixel_t >::type bgr8_pixel_t;
rgb8_pixel_t a( 10, 20, 30 ); bgr8_pixel_t b( 30, 20, 10 );
boost::uint8_t red_a = fusion::at_key< gil_3::red_t >( a ); boost::uint8_t red_b = fusion::at_key< gil_3::red_t >( b ); }
reverse_view is a view after all, so you can't instantiate it with ( 30, 20, 10 ). If you wanna define the reverse one, then fusion::map<..in reverse order...>. I'm happy with the way that GIL defines the pixel, what Fusion really shines here is the generalized way to manipulate the data. HTH
I never had the opportunity to get myself familiar with fusion so I started learning it and I'm really amazed how easy the library is to use. I think I'm at a point where I would like to reach to the community for some advice.
The very idea is to use Fusion to manipulate the pixel, not defining the pixel, though you could do it.
I think I must be misunderstood you then. No problem. But would you agree that gil and fusion share at some code which could be eliminated?
I have one file "adapted_pixel.hpp" that adapts GIL pixel (concept) to Fusion Associative Sequence. Though I want to share here but there's another dependency on my other header.
It's fine when you just point out the general idea. What algorithms in fusion do you use that's not available in gil?
reverse_view is a view after all, so you can't instantiate it with ( 30, 20, 10 ). If you wanna define the reverse one, then fusion::map<..in reverse order...>.
So there is no way to define the return type for reverse view?
I'm happy with the way that GIL defines the pixel, what Fusion really shines here is the generalized way to manipulate the data.
Could you hint me some ideas for what fusion provides that's not in gil? Regards, Christian
2013/3/17 Christian Henning
I never had the opportunity to get myself familiar with fusion so I started learning it and I'm really amazed how easy the library is to use. I think I'm at a point where I would like to reach to the community for some advice.
The very idea is to use Fusion to manipulate the pixel, not defining the pixel, though you could do it.
I think I must be misunderstood you then. No problem. But would you agree that gil and fusion share at some code which could be eliminated?
Ideally, static_xxx in GIL could be eliminated.
I have one file "adapted_pixel.hpp" that adapts GIL pixel (concept) to Fusion Associative Sequence. Though I want to share here but there's another dependency on my other header.
It's fine when you just point out the general idea. What algorithms in fusion do you use that's not available in gil?
I prefer Fusion for the extensibility and flexibility in writing my own algorithms. What in GIL is a fixed set, not for extend, while Fusion can play the role of infrastructure.
reverse_view is a view after all, so you can't instantiate it with ( 30,
20,
10 ). If you wanna define the reverse one, then fusion::map<..in reverse order...>.
So there is no way to define the return type for reverse view?
If you want:
result_of::as_map
I'm happy with the way that GIL defines the pixel, what Fusion really
shines
here is the generalized way to manipulate the data.
Could you hint me some ideas for what fusion provides that's not in gil?
As I said, a generalized way to manipulate the data. You can mix different types (not necessary pixel) to collaborate. And once you developed an algorithm, it's not bound to pixel, but any compatible type. Cheers
participants (2)
-
Christian Henning -
TONGARI