Lubomir,

I'm performing quantitative analysis with a physically significant gray scale.  So I don't expect to do any color conversion.

Thanks for your response.  I'll try out your suggestion when I return from my Christmas vacation.

Curtis Gehman
Burlingame, CA


On Dec 22, 2010, at 2:23 PM, Lubomir Bourdev wrote:

Hi Curtis,

GIL's default float type is not "float" but a special float wrapper to ensure that the range is 0..1
The built-in float has a range from -infinity to infinity and that does not work well when converting a value from another channel type.
If you want an image with a standard float you can create it like this (untested)

typedef pixel<float,gray_layout_t> gray_float_pixel_t;
typedef image<gray_float_pixel_t,false> gray_float_image_t;
typedef gray_float_image_t::view_t gray_float_view_t;
...


However, don't use such images with algorithms that require channel ranges, such as color conversion.

Lubomir


On Dec 20, 2010, at 11:53 PM, Curtis Gehman wrote:

I'm trying to use GIL pixel iterators in a generic STL-style algorithm.  I'm primarily using GIL for the generic 2D iterators.  And in my context, I'm only interested in grayscale images.

The value_type associated with GIL iterators is a pixel type, which typically is an object representing multiple channels (e.g., colors).  It makes sense that comparison operators (e.g., <) are not defined for general color pixel types.

For grayscale images, however, it makes sense that common arithmetic and comparison operators would be defined for the pixel type.  This seems to be the case for the 8-bit grayscale GIL type (gray8_view_t).  The iterator's value_type appears to be seamlessly evaluated to an unsigned char, exactly as I would expect.

When I try to use a float image view type, however, the same magic doesn't seem to happen.  Below is some toy code that demonstrates the issue.  When the last line is knocked out with "//", the code compiles.  But with that second template instantiation (for floats) there, the compiler (gcc 4.2 and VC++2010) exits when it does not find an operator< for the iterator's value_type when instantiating for gray32fc_view_t.

I suppose that I can roll my own less<iterator_traits<SrcView::x_iterator>::value_type> functors.  But I just started using GIL, and I'm hoping that there is a simpler solution.

Any ideas?


#include <algorithm>
#include <iterator>
#include <vector>

#include <boost/gil/gil_all.hpp>

namespace g = boost::gil;

using std::vector;
using std::iterator_traits;

template <typename SrcView> inline
vector<typename iterator_traits<typename SrcView::x_iterator>::value_type>
row_min(const SrcView& src)
{
typedef typename iterator_traits<typename SrcView::x_iterator>::value_type value_t;
vector<value_t> result;
for (int y = 0; y < src.height(); ++y)
{
result.push_back(*std::min_element(src.row_begin(y), src.row_end(y)));
}
return result;
}


// Explicit template instantiation

// Works for gray8c_view_t.
template vector<iterator_traits<g::gray8c_view_t::x_iterator>::value_type> row_min(const g::gray8c_view_t& src);

// Compile error for gray32fc_view_t.
// No operator< for iterator_traits<g::gray32fc_view_t::x_iterator>::value_type.
template vector<iterator_traits<g::gray32fc_view_t::x_iterator>::value_type> row_min(const g::gray32fc_view_t& src);

<ATT00001..txt>

_______________________________________________
Boost-users mailing list
Boost-users@lists.boost.org
http://lists.boost.org/mailman/listinfo.cgi/boost-users