#!python
#!/usr/bin/env python -tt
# encoding: utf-8
#
# Created by Holger Rapp on 2009-03-11.
# HolgerRapp@gmx.net
#
import pycuda.driver as cuda
import pycuda.compiler
import pycuda.autoinit
import numpy
from math import pi,cos,sin
_rotation_kernel_source = """
texture<float, 2> tex;
__global__ void copy_texture_kernel(
const float resize_val,
const float alpha,
unsigned short oldiw, unsigned short oldih,
unsigned short newiw, unsigned short newih,
unsigned char* data) {
// calculate pixel idx
unsigned int x = blockIdx.x * blockDim.x + threadIdx.x;
unsigned int y = blockIdx.y * blockDim.y + threadIdx.y;
// We might be outside the reachable pixels. Don't do anything
if( (x >= newiw) || (y >= newih) )
return;
// calculate offset into destination array
unsigned int didx = y * newiw + x;
// calculate offset into source array (be aware of rotation and scaling)
float xmiddle = (x-newiw/2.) / resize_val;
float ymiddle = (y-newih/2.) / resize_val;
float sx = ( xmiddle*cos(alpha)+ymiddle*sin(alpha) + oldiw/2.) ;
float sy = ( -xmiddle*sin(alpha)+ymiddle*cos(alpha) + oldih/2.);
if( (sx < 0) || (sx >= oldiw) || (sy < 0) || (sy >= oldih) ) {
data[didx] = 255;
return;
}
data[didx] = tex2D(tex, sx, sy);
}
"""
mod_copy_texture=pycuda.compiler.SourceModule( _rotation_kernel_source )
copy_texture_func = mod_copy_texture.get_function("copy_texture_kernel")
texref = mod_copy_texture.get_texref("tex")
def rotate_image( a, resize = 1.5, angle = 20., interpolation = "linear", blocks = (16,16,1) ):
"""
Rotates the array. The new array has the new size and centers the
picture in the middle.
a - array (2-dim)
resize - new_image w/old_image w
angle - degrees to rotate the image
interpolation - "linear" or None
blocks - given to the kernel when run
returns: a new array with dtype=uint8 containing the rotated image
"""
angle = angle/180. *pi
# Convert this image to float. Unsigned int texture gave
# strange results for me. This conversion is slow though :(
a = a.astype("float32")
# Calculate the dimensions of the new image
calc_x = lambda (x,y): (x*a.shape[1]/2.*cos(angle)-y*a.shape[0]/2.*sin(angle))
calc_y = lambda (x,y): (x*a.shape[1]/2.*sin(angle)+y*a.shape[0]/2.*cos(angle))
xs = [ calc_x(p) for p in [ (-1.,-1.),(1.,-1.),(1.,1.),(-1.,1.) ] ]
ys = [ calc_y(p) for p in [ (-1.,-1.),(1.,-1.),(1.,1.),(-1.,1.) ] ]
new_image_dim = (
int(numpy.ceil(max(ys)-min(ys))*resize),
int(numpy.ceil(max(xs)-min(xs))*resize),
)
# Now generate the cuda texture
cuda.matrix_to_texref(a, texref, order="C")
# We could set the next if we wanted to address the image
# in normalized coordinates ( 0 <= coordinate < 1.)
# texref.set_flags(cuda.TRSF_NORMALIZED_COORDINATES)
if interpolation == "linear":
texref.set_filter_mode(cuda.filter_mode.LINEAR)
# Calculate the gridsize. This is entirely given by the size of our image.
gridx = new_image_dim[0]/blocks[0] if \
new_image_dim[0]%blocks[0]==1 else new_image_dim[0]/blocks[0] +1
gridy = new_image_dim[1]/blocks[1] if \
new_image_dim[1]%blocks[1]==0 else new_image_dim[1]/blocks[1] +1
# Get the output image
output = numpy.zeros(new_image_dim,dtype="uint8")
# Call the kernel
copy_texture_func(
numpy.float32(resize), numpy.float32(angle),
numpy.uint16(a.shape[1]), numpy.uint16(a.shape[0]),
numpy.uint16(new_image_dim[1]), numpy.uint16(new_image_dim[0]),
cuda.Out(output),texrefs=[texref],block=blocks,grid=(gridx,gridy))
return output
if __name__ == '__main__':
import Image
import sys
def main( ):
if len(sys.argv) != 2:
print "You should really read the source...\n\nUsage: rotate.py <Imagename>\n"
sys.exit(-1)
# Open, convert to grayscale, convert to numpy array
img = Image.open(sys.argv[1]).convert("L")
i = numpy.fromstring(img.tostring(),dtype="uint8").reshape(img.size[1],img.size[0])
# Rotate & convert back to PIL Image
irot = rotate_image(i)
rotimg = Image.fromarray(irot,mode="L")
# Save and display
rotimg.save("rotated.png")
rotimg.show()
main()