Python transform.rotate函数代码示例

    def rotate(self, img, label, angle):
        assert img.shape==label.shape
        
        img =   t.rotate(img, angle, resize=True)
        label = t.rotate(label, angle, resize=True)
        assert img.shape==label.shape

        lower_x, lower_y = 0 , 0
        while (img[lower_x,:]==0).all(): 
            lower_x +=1    
        while (img[:, lower_y]==0).all():
            lower_y +=1

        upper_x, upper_y = img.shape
        upper_x -=1
        upper_y -=1
        while (img[upper_x,:]==0).all():
            upper_x -=1
        while (img[:, upper_y]==0).all():
            upper_y -=1

        img = img[lower_x:upper_x, lower_y:upper_y]
        label = label[lower_x:upper_x, lower_y:upper_y]

        return img, label

    def loadData(self):
        self.rawData = io.imread(self.fileName, plugin='tifffile')
        self.rawData = cv2.merge((self.rawData[:, :, 0].T,
                                  self.rawData[:, :, 1].T,
                                  self.rawData[:, :, 2].T))
        self.cData = self.rawData.copy()
        self.grayData = self.rawData.copy()
        self.grayData = color.rgb2gray(self.rawData)
        self.hsvData = color.rgb2hsv(self.rawData)
        # self.grayData = self.grayData.convert('LA')
        # self.grayData = self.grayData.transpose(method=PIL.Image.TRANSPOSE)
        self.grayData = transform.rotate(self.grayData, angle=0)
        self.cData = transform.rotate(self.cData, angle=0)
        self.hsvData = transform.rotate(self.hsvData, angle=0)
        self.b = self.cData[:, :, 0]
        self.g = self.cData[:, :, 1]
        self.r = self.cData[:, :, 2]
        self.v = self.hsvData[:, :, 0]
        self.s = self.hsvData[:, :, 1]
        self.h = self.hsvData[:, :, 2]

        self.colorDict = {'RGB': self.cData,
                          'GRAY': self.grayData,
                          'B': self.b,
                          'G': self.g,
                          'R': self.r,
                          'HSV': self.hsvData,
                          'H': self.h,
                          'S': self.s,
                          'V': self.v}

def augmentation(image, imageB, org_width=160,org_height=224, width=190, height=262):
    max_angle=20
    image=cv2.resize(image,(height,width))
    imageB=cv2.resize(imageB,(height,width))

    angle=np.random.randint(max_angle)
    if np.random.randint(2):
        angle=-angle
    image=rotate(image,angle,resize=True)
    imageB=rotate(imageB,angle,resize=True)

    xstart=np.random.randint(width-org_width)
    ystart=np.random.randint(height-org_height)
    image=image[xstart:xstart+org_width,ystart:ystart+org_height]
    imageB=imageB[xstart:xstart+org_width,ystart:ystart+org_height]

    if np.random.randint(2):
        image=cv2.flip(image,1)
        imageB=cv2.flip(imageB,1)
    
    if np.random.randint(2):
        image=cv2.flip(image,0)
        imageB=cv2.flip(imageB,0)

    image=cv2.resize(image,(org_height,org_width))
    imageB=cv2.resize(imageB,(org_height,org_width))

    return image,imageB

	def _augment(self,img, hm, max_rotation = 30):
		""" # TODO : IMPLEMENT DATA AUGMENTATION 
		"""
		if random.choice([0,1]): 
			r_angle = np.random.randint(-1*max_rotation, max_rotation)
			img = 	transform.rotate(img, r_angle, preserve_range = True)
			hm = transform.rotate(hm, r_angle)
		return img, hm

 def iterate_train(self,batchsize,data_augment=False):
     num_batch=40000
     for i in range(num_batch/batchsize):
         start=i*batchsize
         end=(i+1)*batchsize
         if (data_augment==False):
             x=self.train_set_x.get_value(borrow=True)[start:end]
             x=(x-self.mean)/256.0
             x=np.asarray(x,dtype=theano.config.floatX)
             yield x, self.train_set_y.eval()[start:end]
         else:
             imgs=self.train_set_x.get_value(borrow=True)[start:end]
             for j in range(imgs.shape[0]):
                 #horizontally flip
                 if randint(0,1)==0:
                     target=np.copy(imgs[j])
                     for i in range(imgs[j].shape[2]):
                         target[:,:,i]=imgs[j][:,:,imgs[j].shape[2]-1-i]
                     imgs[j]=target
                     
                     
                 #color transform
                 target=np.zeros([3,32,32])
                 mix=range(3)
                 np.random.shuffle(mix)
                 for x in range(3):
                     target[x]=imgs[j][mix[x]]
                 imgs[j]=target
                 
                 
                 r=randint(0,7)
                 if r==0:
                     tmp=np.transpose(imgs[j],(1,2,0));
                     tmp=transform.resize(tmp[0:28,0:28,:],[32,32,3]) 
                     imgs[j]=np.transpose(tmp,(2,0,1))                        
                 elif r==1:
                     tmp=np.transpose(imgs[j],(1,2,0))
                     tmp=transform.resize(tmp[0:28,4:32,:],[32,32,3]) 
                     imgs[j]=np.transpose(tmp,(2,0,1))
                 elif r==2:
                     tmp=np.transpose(imgs[j],(1,2,0))
                     tmp=transform.resize(tmp[4:32,0:28,:],[32,32,3]) 
                     imgs[j]=np.transpose(tmp,(2,0,1))
                 elif r==3:
                     tmp=np.transpose(imgs[j],(1,2,0))
                     tmp=transform.resize(tmp[4:32,4:32,:],[32,32,3]) 
                     imgs[j]=np.transpose(tmp,(2,0,1))
                 elif r==4:
                     tmp=np.asarray(imgs[j],dtype='int32')
                     tmp=transform.rotate(image=tmp,angle=5)
                     imgs[j]=np.asarray(imgs[j],dtype=theano.config.floatX)
                 elif r==5:
                     tmp=np.asarray(imgs[j],dtype='int32')
                     tmp=transform.rotate(image=tmp,angle=-5)
                     imgs[j]=np.asarray(imgs[j],dtype=theano.config.floatX)
             imgs=(imgs-self.mean)/256.0
             imgs=np.asarray(imgs,dtype=theano.config.floatX)
             yield imgs,self.train_set_y.eval()[start:end]

def transform_rot(image):
    
    # Need black background (0 boundary condition) 
    # for rotational transform
    # Thats why Sobel or inverse transformation here
    #image_ref = filters.sobel(image)
    image_ref =  1.0 - image
   
    # Center of mass to be used as rotation center
    m = moments(image_ref,order=1)    
    cx = m[1, 0]/m[0, 0]
    cy = m[0, 1]/m[0, 0]
    com = cx, cy
   
    # This next step is perfect in the math but the rotation angle
    # it generates varies drastically with changes in the watch image
    # thus its not robust enough for universal alignment.
    # Therefore we add an extra rotation step after it.
    # Ascertaining rotation angle from FFT transform
    ind1 = np.arange(image.shape[0],dtype=float)
    ind2 = np.arange(image.shape[1],dtype=float)[:,None]
    angle = \
    np.angle(ind1-com[0]+1j*(ind2-com[1]))
    exp_theta = np.exp(1j*angle)
    angle_rot = np.angle(np.sum(np.sum(image_ref*exp_theta,axis=1)),deg=True)
    # Creating temporary rotated version of input image 
    image_rot_aux = \
    transform.rotate(image,angle_rot,resize=False,center=com,mode='nearest')

    # Second rotation step based on Inertia tensor
    # Again need 0 boundary condition away from object and
    # thus Sobel or inverse transform
    #image_ref = filters.sobel(image_rot_aux)
    image_ref =  1.0 - image_rot_aux

    m = moments(image_ref,order=2)
    Ixx = m[2, 0]/m[0, 0] - np.power(cx,2)
    Iyy = m[0, 2]/m[0, 0] - np.power(cy,2)
    Ixy = m[1, 1]/m[0, 0] - cx*cy
    inertia = [[Ixx, Ixy],[Ixy, Iyy]]
    w, v = np.linalg.eig(inertia)
    idx = w.argsort()[::-1]   
    w = w[idx]
    v = v[:,idx]
    cross = np.cross(v[:,0],v[:,1])
    # Ensuring eigenvectors satisfy right-hand rule
    if (cross < 0):
       v[:,1] *= -1
    
    # Ascertaining rotation angle from inertia tensor eigenvectors
    angle_rad = np.arctan2(v[1,0],v[0,0]) + np.pi/2
    angle_rot = np.degrees(angle_rad)
    
    # Creating final rotated version of input image
    image_rot = \
    transform.rotate(image_rot_aux,angle_rot,resize=False,center=com,mode='nearest')

    return image_rot

def rotate_3d_ski(im, gt):
	im = np.transpose(im, (1, 2, 0))
	gt = np.transpose(gt, (1, 2, 0))
	
	ang = np.random.uniform(0, 360)
	r_im = rotate(im , ang, order=3)
	r_gt = rotate(gt, ang, order=3)
	
	return np.transpose(r_im, (2, 0, 1)), np.transpose(r_gt, (2, 0, 1))

def test_rotate_resize():
    x = np.zeros((10, 10), dtype=np.double)

    x45 = rotate(x, 45, resize=False)
    assert x45.shape == (10, 10)

    x45 = rotate(x, 45, resize=True)
    # new dimension should be d = sqrt(2 * (10/2)^2)
    assert x45.shape == (14, 14)

def test_rotate_center():
    x = np.zeros((10, 10), dtype=np.double)
    x[4, 4] = 1
    refx = np.zeros((10, 10), dtype=np.double)
    refx[2, 5] = 1
    x20 = rotate(x, 20, order=0, center=(0, 0))
    assert_almost_equal(x20, refx)
    x0 = rotate(x20, -20, order=0, center=(0, 0))
    assert_almost_equal(x0, x)

def img_rotate(img, rotate, corner_deg_chance):
    rot_chance = np.random.random()
    if rot_chance < corner_deg_chance:
        return tf.rotate(img, 90)
    if corner_deg_chance <= rot_chance < (corner_deg_chance * 2):
        return tf.rotate(img, 180)
    if (corner_deg_chance * 2) <= rot_chance < (corner_deg_chance * 3):
        return tf.rotate(img, 270)
    return tf.rotate(img, rotate)

def rotate_patch(patch, angle):
    """

    :param patch: patch of size (4, 33, 33)
    :param angle: says how much rotation must be applied
    :return: rotate_patch
    """

    return np.array([rotate(patch[0], angle, resize=False),
                     rotate(patch[1], angle, resize=False),
                     rotate(patch[2], angle, resize=False),
                     rotate(patch[3], angle, resize=False)])

def shape_symmetry(image, center, major_axis, attrs={}, debug=False):
    # pad to make image center coincide with symmetry center
    lesion_mask, _ = pad_for_rotation(image[..., 3], center)

    rotated = rotate(lesion_mask, 90-major_axis.angle)
    flipped = rotated[:,::-1]
    diff = np.logical_xor(rotated, flipped)

    pixels_diff = diff.sum() / 2.
    major_ratio = pixels_diff / rotated.sum()

    if debug:
        print """\
==== Shape Symmetry ====
--- Major Axis ---
num of pixels   : %d
shape sym ratio : %.3f
""" % (pixels_diff, major_ratio)

        plt.subplot(231)
        plt.imshow(rotated)
        plt.subplot(232)
        plt.imshow(flipped)
        plt.subplot(233)
        plt.imshow(diff)

    rotated = rotate(lesion_mask, 180-major_axis.angle)
    flipped = rotated[:,::-1]
    diff = np.logical_xor(rotated, flipped)

    pixels_diff = diff.sum() / 2.
    minor_ratio = pixels_diff / rotated.sum()

    if debug:
        print """\
--- Minor Axis ---
num of pixels   : %d
shape sym ratio : %.3f
""" % (pixels_diff, minor_ratio)

        plt.subplot(234)
        plt.imshow(rotated)
        plt.subplot(235)
        plt.imshow(flipped)
        plt.subplot(236)
        plt.imshow(diff)
        plt.show()

    attrs.update([
        ('Shape Asymmetry Major Ratio', major_ratio),
        ('Shape Asymmetry Minor Ratio', minor_ratio),
        ('--Shape Asymmetry Score', (major_ratio > 0.13)*1 + (minor_ratio > 0.15)*1),
    ])

def _argmin_tilt(tilts, img0, flipped_img180, differ):
    nrows, ncols = img0.shape
    borderY, borderX = nrows//20, ncols//20
    from skimage.transform import rotate
    diffs = []
    for tilt in tilts:
        a = rotate(img0/np.max(img0), -tilt)[borderY:-borderY, borderX:-borderX]
        b = rotate(flipped_img180/np.max(flipped_img180), tilt)[borderY:-borderY, borderX:-borderX]
        diff = differ(a,b)
        print("* tilt=%s, diff=%s" % (tilt, diff))
        diffs.append(diff)
        continue
    return tilts[np.argmin(diffs)]

def get_rotated_sample(X, y, n):
	subset = np.random.random_integers(0, X.shape[0]-1, n)
	X_sub = X[subset]
	for index, img in enumerate(X_sub):
		if index%500==0:
			print "Processed Rotated {}".format(index)
		img1 = tf.rotate(img, np.random.uniform(5,15))
		img1 = img1.reshape(-1, 1, 28, 28)
		img2 = tf.rotate(img, -np.random.uniform(5,15))
		img2 = img2.reshape(-1, 1, 28, 28)
		X = np.vstack( (X, img1) )
		X = np.vstack( (X, img2) )
		y = np.append(y,y[subset[index]])
		y = np.append(y,y[subset[index]])
	return X, y

 def augment_data(self, image, target):
     images = [image.ravel(), ]
     targets = [target, ]
     image_modifiers = (
         lambda x: rotate(x, 90),
         lambda x: rotate(x, 180),
         lambda x: rotate(x, 270),
         lambda x: rotate(x, 45),
         lambda x: swirl(x)
     )
     for i in xrange(self.augmentation):
         img = image_modifiers[i](image)
         images.append(img.ravel())
         targets.append(target)
     return images, targets