Skip to content

D
dlib
Commit 306c9c5b authored by Davis King's avatar Davis King
Browse files
Options

Made spatially_filter_image_separable() use SIMD instructions when filtering 

float data.
parent 9b9ffcba
Hide whitespace changes
Inline Side-by-side
Showing with 310 additions and 82 deletions
dlib/image_transforms/spatial_filtering.h
View file @ 306c9c5b
......@@ -10,6 +10,7 @@
#include "../array2d.h"
#include "../matrix.h"
#include "../geometry/border_enumerator.h"
#include "../simd.h"
#include <limits>
namespace dlib
......@@ -208,119 +209,277 @@ namespace dlib
// ----------------------------------------------------------------------------------------
template <
typename in_image_type,
typename out_image_type,
typename EXP1,
typename EXP2,
typename T
>
typename enable_if_c<pixel_traits<typename out_image_type::type>::grayscale,rectangle>::type
spatially_filter_image_separable (
const in_image_type& in_img,
out_image_type& out_img,
const matrix_exp<EXP1>& row_filter,
const matrix_exp<EXP2>& col_filter,
T scale,
bool use_abs = false,
bool add_to = false
)
namespace impl
{
COMPILE_TIME_ASSERT( pixel_traits<typename in_image_type::type>::has_alpha == false );
COMPILE_TIME_ASSERT( pixel_traits<typename out_image_type::type>::has_alpha == false );
DLIB_ASSERT(scale != 0 && row_filter.size() != 0 && col_filter.size() != 0 &&
is_vector(row_filter) &&
is_vector(col_filter),
"\tvoid spatially_filter_image_separable()"
<< "\n\t Invalid inputs were given to this function."
<< "\n\t scale: "<< scale
<< "\n\t row_filter.size(): "<< row_filter.size()
<< "\n\t col_filter.size(): "<< col_filter.size()
<< "\n\t is_vector(row_filter): "<< is_vector(row_filter)
<< "\n\t is_vector(col_filter): "<< is_vector(col_filter)
template <
typename in_image_type,
typename out_image_type,
typename EXP1,
typename EXP2,
typename T
>
rectangle grayscale_spatially_filter_image_separable (
const in_image_type& in_img,
out_image_type& out_img,
const matrix_exp<EXP1>& _row_filter,
const matrix_exp<EXP2>& _col_filter,
T scale,
bool use_abs,
bool add_to
)
{
const_temp_matrix<EXP1> row_filter(_row_filter);
const_temp_matrix<EXP2> col_filter(_col_filter);
COMPILE_TIME_ASSERT( pixel_traits<typename in_image_type::type>::has_alpha == false );
COMPILE_TIME_ASSERT( pixel_traits<typename out_image_type::type>::has_alpha == false );
DLIB_ASSERT(scale != 0 && row_filter.size() != 0 && col_filter.size() != 0 &&
is_vector(row_filter) &&
is_vector(col_filter),
"\tvoid spatially_filter_image_separable()"
<< "\n\t Invalid inputs were given to this function."
<< "\n\t scale: "<< scale
<< "\n\t row_filter.size(): "<< row_filter.size()
<< "\n\t col_filter.size(): "<< col_filter.size()
<< "\n\t is_vector(row_filter): "<< is_vector(row_filter)
<< "\n\t is_vector(col_filter): "<< is_vector(col_filter)
);
DLIB_ASSERT(is_same_object(in_img, out_img) == false,
"\tvoid spatially_filter_image_separable()"
<< "\n\tYou must give two different image objects"
DLIB_ASSERT(is_same_object(in_img, out_img) == false,
"\tvoid spatially_filter_image_separable()"
<< "\n\tYou must give two different image objects"
);
// if there isn't any input image then don't do anything
if (in_img.size() == 0)
{
out_img.clear();
return rectangle();
}
// if there isn't any input image then don't do anything
if (in_img.size() == 0)
{
out_img.clear();
return rectangle();
}
out_img.set_size(in_img.nr(),in_img.nc());
out_img.set_size(in_img.nr(),in_img.nc());
// figure out the range that we should apply the filter to
const long first_row = (col_filter.size()-1)/2;
const long first_col = (row_filter.size()-1)/2;
const long last_row = in_img.nr() - (col_filter.size()/2);
const long last_col = in_img.nc() - (row_filter.size()/2);
// figure out the range that we should apply the filter to
const long first_row = (col_filter.size()-1)/2;
const long first_col = (row_filter.size()-1)/2;
const long last_row = in_img.nr() - (col_filter.size()/2);
const long last_col = in_img.nc() - (row_filter.size()/2);
const rectangle non_border = rectangle(first_col, first_row, last_col-1, last_row-1);
if (!add_to)
zero_border_pixels(out_img, non_border);
const rectangle non_border = rectangle(first_col, first_row, last_col-1, last_row-1);
if (!add_to)
zero_border_pixels(out_img, non_border);
typedef typename out_image_type::mem_manager_type mem_manager_type;
typedef typename EXP1::type ptype;
typedef typename out_image_type::mem_manager_type mem_manager_type;
typedef typename EXP1::type ptype;
array2d<ptype,mem_manager_type> temp_img;
temp_img.set_size(in_img.nr(), in_img.nc());
array2d<ptype,mem_manager_type> temp_img;
temp_img.set_size(in_img.nr(), in_img.nc());
// apply the row filter
for (long r = 0; r < in_img.nr(); ++r)
{
for (long c = first_col; c < last_col; ++c)
// apply the row filter
for (long r = 0; r < in_img.nr(); ++r)
{
ptype p;
ptype temp = 0;
for (long n = 0; n < row_filter.size(); ++n)
for (long c = first_col; c < last_col; ++c)
{
// pull out the current pixel and put it into p
p = get_pixel_intensity(in_img[r][c-first_col+n]);
temp += p*row_filter(n);
ptype p;
ptype temp = 0;
for (long n = 0; n < row_filter.size(); ++n)
{
// pull out the current pixel and put it into p
p = get_pixel_intensity(in_img[r][c-first_col+n]);
temp += p*row_filter(n);
}
temp_img[r][c] = temp;
}
temp_img[r][c] = temp;
}
}
// apply the column filter
for (long r = first_row; r < last_row; ++r)
{
for (long c = first_col; c < last_col; ++c)
// apply the column filter
for (long r = first_row; r < last_row; ++r)
{
ptype temp = 0;
for (long m = 0; m < col_filter.size(); ++m)
for (long c = first_col; c < last_col; ++c)
{
temp += temp_img[r-first_row+m][c]*col_filter(m);
ptype temp = 0;
for (long m = 0; m < col_filter.size(); ++m)
{
temp += temp_img[r-first_row+m][c]*col_filter(m);
}
temp /= scale;
if (use_abs && temp < 0)
{
temp = -temp;
}
// save this pixel to the output image
if (add_to == false)
{
assign_pixel(out_img[r][c], temp);
}
else
{
assign_pixel(out_img[r][c], temp + out_img[r][c]);
}
}
}
return non_border;
}
temp /= scale;
// ------------------------------------------------------------------------------------
// This overload is optimized to use SIMD instructions when filtering float images with
// float filters.
template <
typename in_image_type,
typename out_image_type,
typename EXP1,
typename EXP2
>
rectangle float_spatially_filter_image_separable (
const in_image_type& in_img,
out_image_type& out_img,
const matrix_exp<EXP1>& _row_filter,
const matrix_exp<EXP2>& _col_filter,
bool add_to
)
{
const_temp_matrix<EXP1> row_filter(_row_filter);
const_temp_matrix<EXP2> col_filter(_col_filter);
DLIB_ASSERT(row_filter.size() != 0 && col_filter.size() != 0 &&
is_vector(row_filter) &&
is_vector(col_filter),
"\tvoid spatially_filter_image_separable()"
<< "\n\t Invalid inputs were given to this function."
<< "\n\t row_filter.size(): "<< row_filter.size()
<< "\n\t col_filter.size(): "<< col_filter.size()
<< "\n\t is_vector(row_filter): "<< is_vector(row_filter)
<< "\n\t is_vector(col_filter): "<< is_vector(col_filter)
);
DLIB_ASSERT(is_same_object(in_img, out_img) == false,
"\tvoid spatially_filter_image_separable()"
<< "\n\tYou must give two different image objects"
);
if (use_abs && temp < 0)
// if there isn't any input image then don't do anything
if (in_img.size() == 0)
{
out_img.clear();
return rectangle();
}
out_img.set_size(in_img.nr(),in_img.nc());
// figure out the range that we should apply the filter to
const long first_row = (col_filter.size()-1)/2;
const long first_col = (row_filter.size()-1)/2;
const long last_row = in_img.nr() - (col_filter.size()/2);
const long last_col = in_img.nc() - (row_filter.size()/2);
const rectangle non_border = rectangle(first_col, first_row, last_col-1, last_row-1);
if (!add_to)
zero_border_pixels(out_img, non_border);
typedef typename in_image_type::mem_manager_type mem_manager_type;
array2d<float,mem_manager_type> temp_img;
temp_img.set_size(in_img.nr(), in_img.nc());
// apply the row filter
for (long r = 0; r < in_img.nr(); ++r)
{
long c = first_col;
for (; c < last_col-3; c+=4)
{
temp = -temp;
simd4f p, temp = 0;
for (long n = 0; n < row_filter.size(); ++n)
{
// pull out the current pixel and put it into p
p.load(&in_img[r][c-first_col+n]);
temp += p*row_filter(n);
}
temp.store(&temp_img[r][c]);
}
for (; c < last_col; ++c)
{
float p;
float temp = 0;
for (long n = 0; n < row_filter.size(); ++n)
{
// pull out the current pixel and put it into p
p = in_img[r][c-first_col+n];
temp += p*row_filter(n);
}
temp_img[r][c] = temp;
}
}
// save this pixel to the output image
if (add_to == false)
// apply the column filter
for (long r = first_row; r < last_row; ++r)
{
long c = first_col;
for (; c < last_col-3; c+=4)
{
assign_pixel(out_img[r][c], temp);
simd4f p, temp = 0;
for (long m = 0; m < col_filter.size(); ++m)
{
p.load(&temp_img[r-first_row+m][c]);
temp += p*col_filter(m);
}
// save this pixel to the output image
if (add_to == false)
{
temp.store(&out_img[r][c]);
}
else
{
p.load(&out_img[r][c]);
temp += p;
temp.store(&out_img[r][c]);
}
}
else
for (; c < last_col; ++c)
{
assign_pixel(out_img[r][c], temp + out_img[r][c]);
float temp = 0;
for (long m = 0; m < col_filter.size(); ++m)
{
temp += temp_img[r-first_row+m][c]*col_filter(m);
}
// save this pixel to the output image
if (add_to == false)
{
out_img[r][c] = temp;
}
else
{
out_img[r][c] += temp;
}
}
}
return non_border;
}
return non_border;
}
} // namespace impl
// ----------------------------------------------------------------------------------------
template <
typename in_image_type,
typename out_image_type,
typename EXP1,
typename EXP2
>
struct is_float_filtering
{
const static bool value = is_same_type<typename in_image_type::type,float>::value &&
is_same_type<typename out_image_type::type,float>::value &&
is_same_type<typename EXP1::type,float>::value &&
is_same_type<typename EXP2::type,float>::value;
};
// ----------------------------------------------------------------------------------------
......@@ -331,15 +490,77 @@ namespace dlib
typename EXP2,
typename T
>
typename disable_if_c<pixel_traits<typename out_image_type::type>::grayscale,rectangle>::type
typename enable_if_c<pixel_traits<typename out_image_type::type>::grayscale &&
is_float_filtering<in_image_type,out_image_type,EXP1,EXP2>::value,rectangle>::type
spatially_filter_image_separable (
const in_image_type& in_img,
out_image_type& out_img,
const matrix_exp<EXP1>& row_filter,
const matrix_exp<EXP2>& col_filter,
T scale,
bool use_abs = false,
bool add_to = false
)
{
if (use_abs == false)
{
if (scale == 1)
return impl::float_spatially_filter_image_separable(in_img, out_img, row_filter, col_filter, add_to);
else
return impl::float_spatially_filter_image_separable(in_img, out_img, row_filter/scale, col_filter, add_to);
}
else
{
return impl::grayscale_spatially_filter_image_separable(in_img, out_img, row_filter, col_filter, scale, true, add_to);
}
}
// ----------------------------------------------------------------------------------------
template <
typename in_image_type,
typename out_image_type,
typename EXP1,
typename EXP2,
typename T
>
typename enable_if_c<pixel_traits<typename out_image_type::type>::grayscale &&
!is_float_filtering<in_image_type,out_image_type,EXP1,EXP2>::value,rectangle>::type
spatially_filter_image_separable (
const in_image_type& in_img,
out_image_type& out_img,
const matrix_exp<EXP1>& _row_filter,
const matrix_exp<EXP2>& _col_filter,
T scale,
bool use_abs = false,
bool add_to = false
)
{
const_temp_matrix<EXP1> row_filter(_row_filter);
const_temp_matrix<EXP2> col_filter(_col_filter);
return impl::grayscale_spatially_filter_image_separable(in_img,out_img, row_filter, col_filter, scale, use_abs, add_to);
}
// ----------------------------------------------------------------------------------------
template <
typename in_image_type,
typename out_image_type,
typename EXP1,
typename EXP2,
typename T
>
typename disable_if_c<pixel_traits<typename out_image_type::type>::grayscale,rectangle>::type
spatially_filter_image_separable (
const in_image_type& in_img,
out_image_type& out_img,
const matrix_exp<EXP1>& _row_filter,
const matrix_exp<EXP2>& _col_filter,
T scale
)
{
const_temp_matrix<EXP1> row_filter(_row_filter);
const_temp_matrix<EXP2> col_filter(_col_filter);
COMPILE_TIME_ASSERT( pixel_traits<typename in_image_type::type>::has_alpha == false );
COMPILE_TIME_ASSERT( pixel_traits<typename out_image_type::type>::has_alpha == false );
......@@ -428,6 +649,8 @@ namespace dlib
return non_border;
}
// ----------------------------------------------------------------------------------------
template <
typename in_image_type,
typename out_image_type,
......@@ -444,6 +667,8 @@ namespace dlib
return spatially_filter_image_separable(in_img,out_img,row_filter,col_filter,1);
}
// ----------------------------------------------------------------------------------------
// ----------------------------------------------------------------------------------------
// ----------------------------------------------------------------------------------------
template <
......
dlib/image_transforms/spatial_filtering_abstract.h
View file @ 306c9c5b
......@@ -121,6 +121,9 @@ namespace dlib
- #out_img.nr() == in_img.nr()
- returns a rectangle which indicates what pixels in #out_img are considered
non-border pixels and therefore contain output from the filter.
- if (use_abs == false && all images and filers contain float types) then
- This function will use SIMD instructions and is particularly fast. So if
you can use this form of the function it can give a decent speed boost.
!*/
// ----------------------------------------------------------------------------------------
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment