def test_toneMap():
hdr=np.load('hdr.npy')
out=a5.toneMap(hdr, 100, 1.0, useBila=False)
io.imwrite(out, 'tone_map_gauss.png')
out=a5.toneMap(hdr, 100, 2.0, useBila=True)
io.imwrite(out, 'tone_map_bila.png')
def testSergei():
sergeis, sergeiNames = getRawPNGsInDir("data/Sergei/")
scount = 0
for f in sergeis:
io.imwrite(a4.split(f), str('splitSergei'+'%03d'%(scount))+'.png')
io.imwrite(a4.sergeiRGB(f), str('Sergei'+'%03d'%(scount))+'.png')
scount = scount +1
def paintAndOutline(im, texture, N=7000, size=50, noise=0.3, debug=False):
'''
On top of painted images, I implement a Sobel filter
pipeline to outline edges black. This is similar
to the game series 'Borderlands' and its art style.
Output image is a grayscale painted-style image.
Color processing for edge detection requires more
advanced techniques (like the one for Borderlands).
See the company's SIGGRAPH talk about Borderlands's development:
https://www.cs.williams.edu/~morgan/SRG10/borderlands.pptx.zip.
'''
edges = gradientMagnitude(helper.BW(im))
black_edges = 1 - edges
if debug:
# should print a picture where the edges are black
io.imwriteGrey(black_edges, str("paintAndOutlineSobelEdges.png"))
# mix with the original color image, but painted!
out = orientedPaint(im, texture, N, size, noise)
for y in xrange(im.shape[0]):
for x in xrange(im.shape[1]):
if black_edges[y, x] < 0.05:
out[y, x] = 0.0
io.imwrite(out, str("paintAndOutlineEdges.png"))
return out
def writeFrames(video, path):
nFrame=video.shape[0]
for i in xrange(nFrame):
pathi=path+str('%03d'%i)+'.png'
#if i%10==0: print i
io.imwrite(video[i], pathi)
print 'wrote'+path+'\n'
def painterly(im,
texture,
N=10000,
size=50,
noise=01.3,
debug=False,
imname=''):
'''First paints at a coarse scale using all 1's for importance sampling, then paints again at size/4 scale using the sharpness map for importance sampling.'''
out = io.constantIm(im.shape[0], im.shape[1])
outCopy = None
if debug:
outCopy = out.copy()
# first pass
importance_first_pass = np.ones_like(im)
singleScalePaint(im, out, importance_first_pass, texture, size, N, noise)
if debug:
io.imwrite(out, str(imname + "PainterlyFirstPassOnly.png"))
# second pass
importance_second_pass = helper.sharpnessMap(im)
singleScalePaint(im, out, importance_second_pass, texture, size / 4, N,
noise)
if debug:
singleScalePaint(im, outCopy, importance_second_pass, texture,
size / 4, N, noise)
io.imwrite(outCopy, str(imname + "PainterlySecondPassOnly.png"))
return out
def testStitchNVancouver():
#im1=io.imread('vancouverPan/vancouver4.png')
#im2=io.imread('vancouverPan/vancouver3.png')
im1=io.imread('vancouverPan/vancouver2.png')
im2=io.imread('vancouverPan/vancouver1.png')
im3=io.imread('vancouverPan/vancouver0.png')
imList = [im1, im2, im3]#, im4, im5]
pointList1a=[np.array([99, 326, 1], dtype=np.float64), np.array([271, 247, 1], dtype=np.float64), np.array([180, 178, 1], dtype=np.float64), np.array([179, 276, 1], dtype=np.float64)]
pointList2a=[np.array([124, 169, 1], dtype=np.float64), np.array([284, 98, 1], dtype=np.float64), np.array([189, 25, 1], dtype=np.float64), np.array([194, 125, 1], dtype=np.float64)]
listOfPairs2=zip(pointList1a, pointList2a)
pointList1b=[np.array([176, 300, 1], dtype=np.float64), np.array([318, 204, 1], dtype=np.float64), np.array([258, 203, 1], dtype=np.float64), np.array([181, 138, 1], dtype=np.float64)]
pointList2b=[np.array([179, 180, 1], dtype=np.float64), np.array([317, 86, 1], dtype=np.float64), np.array([256, 87, 1], dtype=np.float64), np.array([173, 15, 1], dtype=np.float64)]
listOfPairs1=zip(pointList1b, pointList2b)
#pointList1_3=[np.array([165, 186, 1], dtype=np.float64), np.array([173, 146, 1], dtype=np.float64), np.array([188, 80, 1], dtype=np.float64), np.array([164, 40, 1], dtype=np.float64)]
#pointList2_3=[np.array([153, 298, 1], dtype=np.float64), np.array([162, 253, 1], dtype=np.float64), np.array([178, 188, 1], dtype=np.float64), np.array([156, 151, 1], dtype=np.float64)]
#listOfPairs3=zip(pointList1_3, pointList2_3)
#pointList1_4=[np.array([156, 151, 1], dtype=np.float64), np.array([220, 34, 1], dtype=np.float64), np.array([184, 160, 1], dtype=np.float64), np.array([186, 89, 1], dtype=np.float64)]
#pointList2_4=[np.array([151, 304, 1], dtype=np.float64), np.array([226, 189, 1], dtype=np.float64), np.array([180, 316, 1], dtype=np.float64), np.array([190, 239, 1], dtype=np.float64)]
#listOfPairs4=zip(pointList1_4, pointList2_4)
listOfListOfPairs = [listOfPairs1, listOfPairs2]# listOfPairs4]
#listOfListOfPairs = [listOfPairs]#, listOfPairs2]#, listOfPairs2]#, listOfPairs3, listOfPairs4]
out = a6.stitchN(imList, listOfListOfPairs, 0)
io.imwrite(out, "vancouver_stitchN.png")
def testSingleScale(im, texture, outputName, N=10000, size=50, noise=0.3):
out = np.zeros_like(im)
importance = np.zeros_like(im)
importance[150] = 1
## npr.singleScalePaint(im, out, np.ones_like(im), texture, size, N, noise)
npr.singleScalePaint(im, out, importance, texture, size, N, noise)
io.imwrite(out, str(outputName+"SingleScale"+".png"))
def makeConventionManyPano():
conv1 = io.imread("convention/convention-1.png")
conv2 = io.imread("convention/convention-2.png")
conv3 = io.imread("convention/convention-3.png")
pointList1 = [
np.array([298, 206, 1], dtype=np.float64),
np.array([267, 320, 1], dtype=np.float64),
np.array([170, 325, 1], dtype=np.float64),
np.array([172, 188, 1], dtype=np.float64),
]
pointList2 = [
np.array([309, 70, 1], dtype=np.float64),
np.array([270, 175, 1], dtype=np.float64),
np.array([182, 176, 1], dtype=np.float64),
np.array([179, 42, 1], dtype=np.float64),
]
listOfPairs1 = zip(pointList1, pointList2)
pointList3 = [
np.array([288, 173, 1], dtype=np.float64),
np.array([267, 306, 1], dtype=np.float64),
np.array([219, 306, 1], dtype=np.float64),
np.array([217, 210, 1], dtype=np.float64),
]
pointList4 = [
np.array([298, 15, 1], dtype=np.float64),
np.array([269, 151, 1], dtype=np.float64),
np.array([225, 148, 1], dtype=np.float64),
np.array([221, 55, 1], dtype=np.float64),
]
listOfPairs2 = zip(pointList3, pointList4)
listOfImages = [conv1, conv2, conv3]
listOfListOfPairs = [listOfPairs1, listOfPairs2]
refIndex = 1
out = a6.stitchN(listOfImages, listOfListOfPairs, refIndex)
io.imwrite(out, "MyPanoMany.png")
def testStitchScience():
im1=io.imread('science/science-1.png')
im2=io.imread('science/science-2.png')
pointList1=[np.array([307, 15, 1], dtype=np.float64), np.array([309, 106, 1], dtype=np.float64), np.array([191, 102, 1], dtype=np.float64), np.array([189, 47, 1], dtype=np.float64)]
pointList2=[np.array([299, 214, 1], dtype=np.float64), np.array([299, 304, 1], dtype=np.float64), np.array([182, 292, 1], dtype=np.float64), np.array([183, 236, 1], dtype=np.float64)]
listOfPairs=zip(pointList1, pointList2)
out = a6.stitch(im1, im2, listOfPairs)
io.imwrite(out, "science_stitch.png")
def testStitchMIT():
im1=io.imread('mit1.png')
im2=io.imread('mit0.png')
pointList1=[np.array([196, 245, 1], dtype=np.float64), np.array([250, 320, 1], dtype=np.float64), np.array([138, 306, 1], dtype=np.float64), np.array([113, 260, 1], dtype=np.float64)]
pointList2=[np.array([200, 48, 1], dtype=np.float64), np.array([255, 115, 1], dtype=np.float64), np.array([150, 109, 1], dtype=np.float64), np.array([119, 65, 1], dtype=np.float64)]
listOfPairs=zip(pointList1, pointList2)
out = a6.stitch(im1, im2, listOfPairs)
io.imwrite(out, "MyPano.png")
def testCompositeStata():
im1=io.imread('stata/stata-1.png')
im2=io.imread('stata/stata-2.png')
pointList1=[np.array([209, 218, 1]), np.array([425, 300, 1]), np.array([209, 337, 1]), np.array([396, 336, 1])]
pointList2=[np.array([232, 4, 1]), np.array([465, 62, 1]), np.array([247, 125, 1]), np.array([433, 102, 1])]
listOfPairs=zip(pointList1, pointList2)
out = a6.stitchN([im1, im2], [listOfPairs], 0)
io.imwrite(out, "stata_stitchN.png")
def testStitchFun():
im1=io.imread('fun/room1.png')
im2=io.imread('fun/room2.png')
pointList1=[np.array([327, 258, 1], dtype=np.float64), np.array([75, 437, 1], dtype=np.float64), np.array([224, 364, 1], dtype=np.float64), np.array([423, 449, 1], dtype=np.float64)]
pointList2=[np.array([294, 50, 1], dtype=np.float64), np.array([50, 227, 1], dtype=np.float64), np.array([190, 161, 1], dtype=np.float64), np.array([366, 240, 1], dtype=np.float64)]
listOfPairs=zip(pointList1, pointList2)
out = a6.stitch(im1, im2, listOfPairs)
io.imwrite(out, "MyPano.png")
def test_convolve_gauss():
im=io.imread('pru.png')
gauss3=np.array([[1, 2, 1], [2, 4, 2], [1, 2, 1]])
kernel=gauss3.astype(float)
kernel=kernel/sum(sum(kernel))
out=a3.convolve(im, kernel)
io.imwrite(out, 'my_gaussblur.png')
def testStitchStata():
im1 = io.imread("stata/stata-1.png")
im2 = io.imread("stata/stata-2.png")
pointList1 = [np.array([209, 218, 1]), np.array([425, 300, 1]), np.array([209, 337, 1]), np.array([396, 336, 1])]
pointList2 = [np.array([232, 4, 1]), np.array([465, 62, 1]), np.array([247, 125, 1]), np.array([433, 102, 1])]
listOfPairs = zip(pointList1, pointList2)
out = a6.stitch(im1, im2, listOfPairs)
io.imwrite(out, "stata_stitch.png")
def testAngle(im):
thetas = npr.computeAngles(im)
out = np.zeros_like(thetas)
for y in xrange(thetas.shape[0]):
for x in xrange(thetas.shape[1]):
theta = thetas[y,x,0]
if theta<0: theta+=2*math.pi
out[y,x]=theta/2.0/math.pi
io.imwrite(out, 'testangle.png')
def testCatalina(im, N=50000, size=50, noise=0.2):
'''
Renders a painted version of Catalina.
Refer to the macOS Catalina default dynamic wallpaper
for this.
'''
print 'TEST 4: oriented brushing on macOS Catalina wallpaper'
io.imwrite(npr.orientedPaint(im, brush1, N, size, noise),
str("CatalinaOrientedPaint" + ".png"))
def testAngle(im):
thetas = npr.computeAngles(im)
out = np.zeros_like(thetas)
for y in xrange(thetas.shape[0]):
for x in xrange(thetas.shape[1]):
theta = thetas[y, x, 0]
if theta < 0: theta += 2 * math.pi
out[y, x] = theta / 2.0 / math.pi
io.imwrite(out, 'testangle.png')
def test_3_toneMap(self):
im1 = numpy.load('vine-hdr.npy')
# im2 = numpy.load('design-hdr.npy')
# im3 = numpy.load('ante2-hdr.npy')
t1 = toneMap(im1, 100, 1, False)
# t2 = toneMap(im2)
# t3 = toneMap(im3)
imageIO.imwrite(t1, 'vine-toned-g.png')
def main():
# This program defines a single-stage imaging pipeline that
# brightens an image.
# First we'll load the input image we wish to brighten.
# We'll use imageIO to get a numpy array from a PNG image
im=imageIO.imread('rgb.png')
# We then create a Halide representation of this image using the Image
# constructor
input = Image(Float(32), im)
# the first input to the Image constructor is a type 32-bit float here)
# the second can be a filename, a numpy array or nothing
# when it's a filename, the file gets loaded
# Next we declare our Func object that represents our one pipeline
# stage.
sel=Func()
# Our Func will have three arguments, representing the position
# in the image and the color channel. Halide treats color
# channels as an extra dimension of the image, just like in numpy
# let's declare the corresponding Vars before we can use them
x, y, c = Var(), Var(), Var()
# Finally define the function.
sel[x, y, c] = select(input[x,y,1]<0.5, 0.0, 1.0)
# The equivalent one-liner to all of the above is:
#
# brighter[x, y, c] = input[x, y, c] * 1.5
#
# Remember. All we've done so far is build a representation of a
# Halide program in memory. We haven't actually processed any
# pixels yet. We haven't even compiled that Halide program yet.
# So now we'll realize the Func. The size of the output image
# should match the size of the input image. If we just wanted to
# brighten a portion of the input image we could request a
# smaller size. If we request a larger size Halide will throw an
# error at runtime telling us we're trying to read out of bounds
# on the input image.
output = sel.realize(input.width(), input.height(), input.channels());
# realize provides us with some Halide internal datatype representing image buffers.
# We want to convert it to a numpy array. For this, we first turn it into a
# proper Halide Image using the Halide constructor Image(), and we then convert
# it to a numpy array. It's a little verbose but not a big deal.
outputNP=numpy.array(Image(output))
imageIO.imwrite(outputNP, 'sel.png')
print "Success!\n"
return 0;
def spanish(im):
L = rgb2yuv(im)
u = L[:, :, 1]
L[:, :, 1] = L[:, :, 2]
L[:, :, 2] = u
dotInMiddle(L)
imageIO.imwrite(L, "L.png")
C = BW(im)
dotInMiddle(C)
imageIO.imwrite(C, "C.png")
def main():
# This program defines a multi-stage Halide imaging pipeline
# One stage computes the horixontal gradient of an image dI/dx
# Another stage computes dI/dy (for all three channels of RGB in both cases)
# The final stage computes the magnitude of the corresponding vector
# We will compute the gradient with finite differences: dI/dx=I(x+1)-I(x)
# As usual, let's load an input
im=imageIO.imread('rgb.png')
# and create a Halide representation of this image
input = Image(Float(32), im)
# Next we declaure the Vars
# We here give an extra argument to the Var constructor, an optional string that
# can help for debugging by naming the variable in the Halide representation.
# Otherwise, the names x, y, c are only known to the Python side
x, y, c = Var('x'), Var('y'), Var('c')
# Next we declare the three Funcs corresponding to the various stages of the gradient .
# Similarly, we pass strings to name them.
gx = Func('gx')
gy = Func('gy')
gradientMagnitude=Func('gradientMagnitude')
# Define our horizontal gradient Func using finite difference
# The value at a pixel is the input at that pixel minus its left neighbor.
# Note how we now use the more direct definition of Funcs without declaring
# intermediate Exprs
gx[x,y,c]=input[x+1,y,c]-input[x, y,c]
# Similarly define the vertical gradient.
gy[x,y,c]=input[x,y+1,c]-input[x,y,c]
# Finally define the gradient magnitude as the Euclidean norm of the gradient vector
# We use overloaded operators and functions such as **, + and sqrt
# Through he magic of metaprogramming, this creates the appropriate algebraic tree
# in Halide representation
# Most operators and functions you expect are supported.
# Check the documentation for the full list.
gradientMagnitude[x,y,c]= sqrt(gx[x,y,c]**2+gy[x,y,c]**2)
# As usual, all we have done so far is create a Halide internal representation.
# No computation has happened yet.
# We now call realize() to compile and execute.
# You'll note that we subtracted 1 from the width and height to make sure that the
# x+1 and y+1 neighbors always exist. We'll see a more general solution in the next tutorial
output = gradientMagnitude.realize(input.width()-1, input.height()-1, input.channels());
outputNP=numpy.array(Image(output))
imageIO.imwrite(outputNP, 'tut3out.png', gamma=1.0)
print 'success!'
return 0
def test_makeHDR():
import glob, time
inputs=sorted(glob.glob('data/ante0-*.png'))
p=multi.Pool(processes=8)
im_list=p.map(io.imread,inputs)
hdr=a5.makeHDR(im_list)
np.save('hdr', hdr)
hdr_scale=hdr/max(hdr.flatten())
io.imwrite(hdr_scale, 'hdr_linear_scale_ante0.png')
def testComputeAndApplyHomographyStata():
im1=io.imread('stata/stata-1.png')
im2=io.imread('stata/stata-2.png')
pointList1=[np.array([209, 218, 1]), np.array([425, 300, 1]), np.array([209, 337, 1]), np.array([396, 336, 1])]
pointList2=[np.array([232, 4, 1]), np.array([465, 62, 1]), np.array([247, 125, 1]), np.array([433, 102, 1])]
listOfPairsS=zip(pointList1, pointList2)
HS=a6.computeHomography(listOfPairsS)
#multiply by 0.2 to better show the transition
out=im2*0.5
a6.applyHomography(im1, out, HS, True)
io.imwrite(out, "stata_computeAndApplyHomography.png")
def testForest(im, N=50000, size=25, noise=0.4):
'''
Tests NPR on a beautiful image of a forest I found
on Google Images.
Generates a squiggly, worm-looking image.
Quite abstract.
'''
print 'TEST 2: oriented brushing on beautiful forest image'
io.imwrite(npr.orientedPaint(im, bigBrush, N, size, noise),
str("ForestOrientedPaint" + ".png"))
def testComputeAndApplyHomographyFun():
im1=io.imread('fun/room1.png')
im2=io.imread('fun/room2.png')
pointList1=[np.array([327, 258, 1], dtype=np.float64), np.array([75, 437, 1], dtype=np.float64), np.array([224, 364, 1], dtype=np.float64), np.array([423, 449, 1], dtype=np.float64)]
pointList2=[np.array([294, 50, 1], dtype=np.float64), np.array([50, 227, 1], dtype=np.float64), np.array([190, 161, 1], dtype=np.float64), np.array([366, 240, 1], dtype=np.float64)]
listOfPairsS=zip(pointList1, pointList2)
HS=a6.computeHomography(listOfPairsS)
#multiply by 0.2 to better show the transition
out=im2*0.5
a6.applyHomography(im1, out, HS, True)
io.imwrite(out, "fun.png")
def main():
im=imageIO.imread('hk.png', 1.0)
t=time.time()
out=harris(im)
dt=time.time()-t
print 'took ', dt, 'seconds'
norm=np.max(out)
imageIO.imwrite(out/norm)
def testApplyHomographyPoster():
signH = np.array(
[
[1.12265192e00, 1.44940136e-01, 1.70000000e02],
[8.65164180e-03, 1.19897030e00, 9.50000000e01],
[2.55704864e-04, 8.06420365e-04, 1.00000000e00],
]
)
green = io.getImage("green.png")
poster = io.getImage("poster.png")
a6.applyHomography(poster, green, signH, True)
io.imwrite(green, "HWDueAt9pm_applyHomography.png")
def test_makeHDR():
import glob, time
inputs=glob.glob('data/sea-*.png')
im_list = []
for inp in inputs:
im_list.append(io.imread(inp))
hdr=a5.makeHDR(im_list)
np.save('hdr', hdr)
hdr_scale=hdr/max(hdr.flatten())
io.imwrite(hdr_scale, 'hdr_linear_scale.png')
def testSingleScaleOrientedPaint(im,
texture,
outputName,
N=10000,
size=50,
noise=0.3,
nAngles=36):
out = np.zeros_like(im)
thetas = npr.computeAngles(im)
npr.singleScaleOrientedPaint(im, out, thetas, np.ones_like(im), texture,
size, N, noise, nAngles)
io.imwrite(out, str(outputName + "SingleScaleOriented" + ".png"))
def main():
im=np.load('Input/hk.npy')
mpix= (im.shape[0] * im.shape[1])/1e6
t=time.time()
out=harris(im)
dt=time.time()-t
print 'took ', dt, 'seconds'
print '%.5f ms per megapixel (%.7f ms for %d megapixels)' % (dt/mpix*1e3, dt*1e3, mpix)
norm=np.max(out)
imageIO.imwrite(out/norm)
def test_makeHDR(file):
import glob
inputs=sorted(glob.glob('data/' + file + '-*.png'))
im_list = []
for inp in inputs:
im_list.append(io.imread(inp))
hdr=a5.makeHDR(im_list)
np.save('npy/'+file+'hdr', hdr)
hdr_scale=hdr/max(hdr.flatten())
io.imwrite(hdr_scale, file+'_hdr_linear_scale.png')
def main():
#im=imageIO.imread('rgb.png')
im=np.load('Input/hk.npy')
t=time.time()
out=harris(im)
dt=time.time()-t
print 'took ', dt, 'seconds'
norm=np.max(out)
imageIO.imwrite(out/norm, "FredoHarris.png")
def main():
im=imageIO.imread('hk.png')
im = im[:im.shape[0]/1.2, :im.shape[1]/1.2]
t=time.time()
out=harris(im)
dt=time.time()-t
print 'took ', dt, 'seconds'
mp = im.shape[0] * im.shape[1] / 1e6
print dt / float(mp), 'seconds / mp'
# norm=np.max(out)
imageIO.imwrite(out)
def testComputeAndApplyHomographyPoster():
green = io.getImage("green.png")
poster = io.getImage("poster.png")
h, w = poster.shape[0] - 1, poster.shape[1] - 1
pointListPoster = [np.array([0, 0, 1]), np.array([0, w, 1]), np.array([h, w, 1]), np.array([h, 0, 1])]
pointListT = [np.array([170, 95, 1]), np.array([171, 238, 1]), np.array([233, 235, 1]), np.array([239, 94, 1])]
listOfPairs = zip(pointListPoster, pointListT)
H = a6.computeHomography(listOfPairs)
# print H
a6.applyHomography(poster, green, H, True)
io.imwrite(green, "HWDueAt9pm_computeHomography.png")
def testNPanoGuedelon():
im1 = io.imread("guedelon/guedelon-1.png")
im2 = io.imread("guedelon/guedelon-2.png")
im3 = io.imread("guedelon/guedelon-3.png")
im4 = io.imread("guedelon/guedelon-4.png")
pointList1 = [
np.array([444, 306, 1], dtype=np.float64),
np.array([198, 210, 1], dtype=np.float64),
np.array([271, 198, 1], dtype=np.float64),
np.array([399, 203, 1], dtype=np.float64),
]
pointList2 = [
np.array([434, 114, 1], dtype=np.float64),
np.array([188, 44, 1], dtype=np.float64),
np.array([261, 24, 1], dtype=np.float64),
np.array([394, 18, 1], dtype=np.float64),
]
listOfPairs1 = zip(pointList1, pointList2)
pointList3 = [
np.array([419, 293, 1], dtype=np.float64),
np.array([384, 234, 1], dtype=np.float64),
np.array([254, 274, 1], dtype=np.float64),
np.array([301, 324, 1], dtype=np.float64),
]
pointList4 = [
np.array([401, 142, 1], dtype=np.float64),
np.array([372, 88, 1], dtype=np.float64),
np.array([245, 139, 1], dtype=np.float64),
np.array([292, 179, 1], dtype=np.float64),
]
listOfPairs2 = zip(pointList3, pointList4)
pointList5 = [
np.array([245, 139, 1], dtype=np.float64),
np.array([403, 143, 1], dtype=np.float64),
np.array([379, 220, 1], dtype=np.float64),
np.array([273, 311, 1], dtype=np.float64),
]
pointList6 = [
np.array([236, 70, 1], dtype=np.float64),
np.array([398, 69, 1], dtype=np.float64),
np.array([371, 149, 1], dtype=np.float64),
np.array([267, 238, 1], dtype=np.float64),
]
listOfPairs3 = zip(pointList5, pointList6)
listOfImages = [im1, im2, im3, im4]
listOfListOfPairs = [listOfPairs1, listOfPairs2, listOfPairs3]
refIndex = 2
out = a6.stitchN(listOfImages, listOfListOfPairs, refIndex)
io.imwrite(out, "guedelon_stitchNFast.png")
def makeStreetSign():
sign = io.imread("highway.png")
people = io.imread("coverphoto.png")
h, w = people.shape[0] - 1, people.shape[1] - 1
peoplecorners = [np.array([0, 0, 1]), np.array([0, w, 1]), np.array([h, w, 1]), np.array([h, 0, 1])]
pointList1 = [
np.array([105, 94, 1], dtype=np.float64),
np.array([110, 200, 1], dtype=np.float64),
np.array([162, 200, 1], dtype=np.float64),
np.array([159, 92, 1], dtype=np.float64),
]
listOfPairs = zip(peoplecorners, pointList1)
H = a6.computeHomography(listOfPairs)
a6.applyHomography(people, sign, H, True)
io.imwrite(sign, "Fun.png")
def writeFrames(video, folder, framerate=30):
'''writes the frames of video to path.'''
# make the folder if it doesn't exist
if not os.path.exists(folder):
os.makedirs(folder)
nFrame = video.shape[0]
for i in tqdm(range(nFrame)):
image_path = os.path.join(folder, "frame_%03d.png" % i)
io.imwrite(video[i], image_path)
print('Saved images to {}'.format(folder))
# now save the video
os.system("ffmpeg -y -r {} -i {} -vcodec mpeg4 {}.mp4".format(
framerate, os.path.join(folder, "frame_%03d.png"), folder))
print("Saved video to {}.mp4".format(folder))
def videoMagPhase(phasePyramid, magnitudePyramid, low, high, order, gains_y, gains_uv=None, frame_write_path=None):
for i in xrange(phasePyramid.shape[1]):
print("Processing layer: " + str(i+1) + "/" + str(phasePyramid.shape[1]))
phases = phasePyramid[:,i,:,:,:]
if gains_uv is not None:
print("\tRGB to YUV")
phases = RGB2YUV(phases)
print("\tButterworth filter")
tbpPhases = timeBandPassButter(phases - phases[0], low, high, order)
print("\tPhase amplification")
diffPhase = np.diff(tbpPhases, n=1, axis=0)
diffPhase = (diffPhase + np.pi) % (2 * np.pi) - np.pi
if gains_uv is not None:
diffPhase[:,:,:,0] *= gains_y[i]
diffPhase[:,:,:,1:] *= gains_uv[i]
else:
diffPhase *= gains_y[i]
tbpPhases[1:,:,:,:] = np.cumsum(diffPhase, axis=0) + tbpPhases[0,:,:,:]
if gains_uv is not None:
print("\tYUV to RGB")
phases= YUV2RGB(phases + tbpPhases)
else:
phases = (phases + tbpPhases)
# print("\tPhase denoising")
# phases = ndimage.filters.median_filter(phases, (1, 3, 3, 1))
# for j in xrange(phases.shape[0]):
# print("\t\tFrame " + str(j+1) + "/" + str(phases.shape[0]))
# phases[j] = bilagrid.bilateral_grid(phases[j], max(phases.shape[1], phases.shape[2])/50.0, 0.4)
phasePyramid[:,i,:,:,:] = phases
print("Reconstructing")
frame = np.zeros((phasePyramid.shape[2], phasePyramid.shape[3], phasePyramid.shape[4]))
if frame_write_path is None:
video = np.zeros((phasePyramid.shape[0], phasePyramid.shape[2], phasePyramid.shape[3], phasePyramid.shape[4]))
for i in xrange(phasePyramid.shape[0]):
print("Reconstructing frame " + str(i+1) + "/" + str(phasePyramid.shape[0]))
for j in xrange(3):
frame[:,:,j] = reconstruct((magnitudePyramid[i,:,:,:,j], phasePyramid[i,:,:,:,j]))
if frame_write_path is not None:
io.imwrite(frame, frame_write_path + str(i).zfill(5) + ".png")
else:
video[i] = frame
if frame_write_path is None:
return video
else:
return None
import a1
import imageIO as io
castle = io.imread('castle_small.png')
imL, imC = a1.spanish(castle)
io.imwrite(imL, 'L.png')
io.imwrite(imC, 'C.png')
def testOrientedPaint(im, texture, outputName, N=10000, size=50, noise=0.3):
io.imwrite(npr.orientedPaint(im, texture, N, size, noise),
str(outputName + "OrientedPaint" + ".png"))
def testPainterly(im, texture, outputName, N=10000, size=50, noise=0.3):
io.imwrite(npr.painterly(im, texture, N, size, noise),
str(outputName + "Painterly" + ".png"))
def testSingleScale(im, texture, outputName, N=10000, size=50, noise=0.3):
out = np.zeros_like(im)
npr.singleScalePaint(im, out, np.ones_like(im), texture, size, N, noise)
io.imwrite(out, str(outputName + "SingleScale" + ".png"))
str(outputName + "OrientedPaint" + ".png"))
def runTests(im, texture, imname):
testSingleScale(im, texture, imname)
testPainterly(im, texture, imname)
testSingleScaleOrientedPaint(im, texture, imname)
testOrientedPaint(im, texture, imname)
brush1 = io.imread('brush.png')
longBrush = io.imread('longBrush.png')
bigBrush = io.imread('longBrush2.png')
roundIm = io.imread('round.png')
liz = io.imread('liz.png')
china = io.imread('china.png')
vpd = io.imread('villeperdue.png')
brushtest = np.zeros([200, 200, 3])
testBrush(brushtest, brush1)
io.imwrite(brushtest, "brushtest1.png")
testAngle(roundIm)
runTests(liz, brush1, "Liz")
runTests(china, brush1, "China")
runTests(vpd, brush1, "VPD")
runTests(liz, longBrush, "LizTestLongBrush")
runTests(china, longBrush, "ChinaTestLongBrush")
runTests(roundIm, longBrush, "RoundImLong")