Related
I have 3d volume. Which has shape of (399 x 512 x 512). And It has voxel spacing of 0.484704 x 0.484704 x 0.4847
Now, I want to define a cylinder inside this volume with length 5mm, diameter 1mm, intensity 1 inside, intensity 0 outside.
I saw an example to define a cylinder in internet like this code:
from mpl_toolkits.mplot3d import Axes3D
def data_for_cylinder_along_z(center_x,center_y,radius,height_z):
z = np.linspace(0, height_z, 50)
theta = np.linspace(0, 2*np.pi, 50)
theta_grid, z_grid=np.meshgrid(theta, z)
x_grid = radius*np.cos(theta_grid) + center_x
y_grid = radius*np.sin(theta_grid) + center_y
return x_grid,y_grid,z_grid
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')
Xc,Yc,Zc = data_for_cylinder_along_z(0.2,0.2,0.05,0.1)
ax.plot_surface(Xc, Yc, Zc, alpha=0.5)
plt.show()
However, I don't know how to define the cylinder inside the 3d volume keeping all the conditions (length 5mm, diameter 1mm, intensity 1 inside, intensity 0 outside) true. I also want to define the center of cylinder automatically. So that I can define the cylinder at any place of inside the 3d volume keeping the other condition true. Can anyone show or provide any example?
Thanks a lot in advance.
One simple way of solving this would be to perform each of the checks individually and then just keep the voxels that satisfy all of your constraints.
If you build a grid with all of the centers of the voxels: P (399 x 512 x 512 x 3), each voxel at (i,j,k) will be associated with its real-world position (x,y,z).
That's a little tricky, but it should look something like this:
np.stack(np.meshgrid(np.arange(0, shape[0]),
np.arange(0, shape[1]),
np.arange(0, shape[2]), indexing='ij'), axis=3)
If you subtract the cylinder's center (center_x,center_y, center_z), you're left with the relative positions of each (i,j,k) voxel P_rel (399 x 512 x 512 x 3)
When you have that, you can apply each of your tests one after the other. For a Z-oriented cylinder with a radius and height_z it would look something like:
# constrain the Z-axis
not_too_high = P_rel[:,:,:,2]<= (0.5*height_z)
not_too_low = P_rel[:,:,:,2]>= (-0.5*height_z)
# constrain the radial direction
not_too_far = np.linalg.norm(P_rel[:,:,:,:2],axis=3)<=radius
voxels_in_cyl = not_too_high & not_too_low & not_too_far
I haven't tested the code, but you get the idea.
If you wanted to have an cylinder with an arbitrary orientation you would have to project P_rel into axial and radial components and then do an analogous check without "hard-coding" the indices as I did in this example
I have 2 sets of datapoints:
import random
import pandas as pd
A = pd.DataFrame({'x':[random.uniform(0, 1) for i in range(0,100)], 'y':[random.uniform(0, 1) for i in range(0,100)]})
B = pd.DataFrame({'x':[random.uniform(0, 1) for i in range(0,100)], 'y':[random.uniform(0, 1) for i in range(0,100)]})
For each one of these dataset I can produce the jointplot like this:
import seaborn as sns
sns.jointplot(x=A["x"], y=A["y"], kind='kde')
sns.jointplot(x=B["x"], y=B["y"], kind='kde')
Is there a way to calculate the "common area" between these 2 joint plots ?
By common area, I mean, if you put one joint plot "inside" the other, what is the total area of intersection. So if you imagine these 2 joint plots as mountains, and you put one mountain inside the other, how much does one fall inside the other ?
EDIT
To make my question more clear:
import matplotlib.pyplot as plt
import scipy.stats as st
def plot_2d_kde(df):
# Extract x and y
x = df['x']
y = df['y']
# Define the borders
deltaX = (max(x) - min(x))/10
deltaY = (max(y) - min(y))/10
xmin = min(x) - deltaX
xmax = max(x) + deltaX
ymin = min(y) - deltaY
ymax = max(y) + deltaY
# Create meshgrid
xx, yy = np.mgrid[xmin:xmax:100j, ymin:ymax:100j]
# We will fit a gaussian kernel using the scipy’s gaussian_kde method
positions = np.vstack([xx.ravel(), yy.ravel()])
values = np.vstack([x, y])
kernel = st.gaussian_kde(values)
f = np.reshape(kernel(positions).T, xx.shape)
fig = plt.figure(figsize=(13, 7))
ax = plt.axes(projection='3d')
surf = ax.plot_surface(xx, yy, f, rstride=1, cstride=1, cmap='coolwarm', edgecolor='none')
ax.set_xlabel('x')
ax.set_ylabel('y')
ax.set_zlabel('PDF')
ax.set_title('Surface plot of Gaussian 2D KDE')
fig.colorbar(surf, shrink=0.5, aspect=5) # add color bar indicating the PDF
ax.view_init(60, 35)
I am interested in finding the interection/common volume (just the number) of these 2 kde plots:
plot_2d_kde(A)
plot_2d_kde(B)
Credits: The code for the kde plots is from here
I believe this is what you're looking for. I'm basically calculating the space (integration) of the intersection (overlay) of the two KDE distributions.
A = pd.DataFrame({'x':[random.uniform(0, 1) for i in range(0,100)], 'y':[random.uniform(0, 1) for i in range(0,100)]})
B = pd.DataFrame({'x':[random.uniform(0, 1) for i in range(0,100)], 'y':[random.uniform(0, 1) for i in range(0,100)]})
# KDE fro both A and B
kde_a = scipy.stats.gaussian_kde([A.x, A.y])
kde_b = scipy.stats.gaussian_kde([B.x, B.y])
min_x = min(A.x.min(), B.x.min())
min_y = min(A.y.min(), B.y.min())
max_x = max(A.x.max(), B.x.max())
max_y = max(A.y.max(), B.y.max())
print(f"x is from {min_x} to {max_x}")
print(f"y is from {min_y} to {max_y}")
x = [a[0] for a in itertools.product(np.arange(min_x, max_x, 0.01), np.arange(min_y, max_y, 0.01))]
y = [a[1] for a in itertools.product(np.arange(min_x, max_x, 0.01), np.arange(min_y, max_y, 0.01))]
# sample across 100x100 points.
a_dist = kde_a([x, y])
b_dist = kde_b([x, y])
print(a_dist.sum() / len(x)) # intergral of A
print(b_dist.sum() / len(x)) # intergral of B
print(np.minimum(a_dist, b_dist).sum() / len(x)) # intergral of the intersection between A and B
The following code compares calculating the volume of the intersection either via scipy's dblquad or via taking the average value over a grid.
Remarks:
For the 2D case (and with only 100 sample points), it seems the delta's need to be quite larger than 10%. The code below uses 25%. With a delta of 10%, the calculated values for f1 and f2 are about 0.90, while in theory they should be 1.0. With a delta of 25%, these values are around 0.994.
To approximate the volume the simple way, the average needs to be multiplied by the area (here (xmax - xmin)*(ymax - ymin)). Also, the more grid points are considered, the better the approximation. The code below uses 1000x1000 grid points.
Scipy has some special functions to calculate the integral, such as scipy.integrate.dblquad. This is much slower than the 'simple' method, but a bit more precise. The default precision didn't work, so the code below reduces that precision considerably. (dblquad outputs two numbers: the approximate integral and an indication of the error. To only get the integral, dblquad()[0] is used in the code.)
The same approach can be used for more dimensions. For the 'simple' method, create a more dimensional grid (xx, yy, zz = np.mgrid[xmin:xmax:100j, ymin:ymax:100j, zmin:zmax:100j]). Note that a subdivision by 1000 in each dimension would create a grid that's too large to work with.
When using scipy.integrate, dblquad needs to be replaced by tplquad for 3 dimensions or nquad for N dimensions. This probably will also be rather slow, so the accuracy needs to be reduced further.
import numpy as np
import pandas as pd
import scipy.stats as st
from scipy.integrate import dblquad
df1 = pd.DataFrame({'x':np.random.uniform(0, 1, 100), 'y':np.random.uniform(0, 1, 100)})
df2 = pd.DataFrame({'x':np.random.uniform(0, 1, 100), 'y':np.random.uniform(0, 1, 100)})
# Extract x and y
x1 = df1['x']
y1 = df1['y']
x2 = df2['x']
y2 = df2['y']
# Define the borders
deltaX = (np.max([x1, x2]) - np.min([x1, x2])) / 4
deltaY = (np.max([y1, y2]) - np.min([y1, y2])) / 4
xmin = np.min([x1, x2]) - deltaX
xmax = np.max([x1, x2]) + deltaX
ymin = np.min([y1, y2]) - deltaY
ymax = np.max([y1, y2]) + deltaY
# fit a gaussian kernel using scipy’s gaussian_kde method
kernel1 = st.gaussian_kde(np.vstack([x1, y1]))
kernel2 = st.gaussian_kde(np.vstack([x2, y2]))
print('volumes via scipy`s dblquad (volume):')
print(' volume_f1 =', dblquad(lambda y, x: kernel1((x, y)), xmin, xmax, ymin, ymax, epsabs=1e-4, epsrel=1e-4)[0])
print(' volume_f2 =', dblquad(lambda y, x: kernel2((x, y)), xmin, xmax, ymin, ymax, epsabs=1e-4, epsrel=1e-4)[0])
print(' volume_intersection =',
dblquad(lambda y, x: np.minimum(kernel1((x, y)), kernel2((x, y))), xmin, xmax, ymin, ymax, epsabs=1e-4, epsrel=1e-4)[0])
Alternatively, one can calculate the mean value over a grid of points, and multiply the result by the area of the grid. Note that np.mgrid is much faster than creating a list via itertools.
# Create meshgrid
xx, yy = np.mgrid[xmin:xmax:1000j, ymin:ymax:1000j]
positions = np.vstack([xx.ravel(), yy.ravel()])
f1 = np.reshape(kernel1(positions).T, xx.shape)
f2 = np.reshape(kernel2(positions).T, xx.shape)
intersection = np.minimum(f1, f2)
print('volumes via the mean value multiplied by the area:')
print(' volume_f1 =', np.sum(f1) / f1.size * ((xmax - xmin)*(ymax - ymin)))
print(' volume_f2 =', np.sum(f2) / f2.size * ((xmax - xmin)*(ymax - ymin)))
print(' volume_intersection =', np.sum(intersection) / intersection.size * ((xmax - xmin)*(ymax - ymin)))
Example output:
volumes via scipy`s dblquad (volume):
volume_f1 = 0.9946974276169385
volume_f2 = 0.9928998852123891
volume_intersection = 0.9046421634401607
volumes via the mean value multiplied by the area:
volume_f1 = 0.9927873844924111
volume_f2 = 0.9910132867915901
volume_intersection = 0.9028999384136771
I need to implement a solver for linear programming problems. All of the restrictions are <= ones such as
5x + 10y <= 10
There can be an arbitrary amount of these restrictions. Also , x>=0 y>=0 implicitly.
I need to find the optimal solutions(max) and show the feasible region in matplotlib. I've found the optimal solution by implementing the simplex method but I can't figure out how to draw the graph.
Some approaches I've found:
This link finds the minimum of the y points from each function and uses plt.fillBetween() to draw the region. But it doesn't work when I change the order of the equations. I'm not sure which y values to minimize(). So I can't use it for arbitrary restrictions.
Find solution for every pair of restrictions and draw a polygon. Not efficient.
An easier approach might be to have matplotlib compute the feasible region on its own (with you only providing the constraints) and then simply overlay the "constraint" lines on top.
# plot the feasible region
d = np.linspace(-2,16,300)
x,y = np.meshgrid(d,d)
plt.imshow( ((y>=2) & (2*y<=25-x) & (4*y>=2*x-8) & (y<=2*x-5)).astype(int) ,
extent=(x.min(),x.max(),y.min(),y.max()),origin="lower", cmap="Greys", alpha = 0.3);
# plot the lines defining the constraints
x = np.linspace(0, 16, 2000)
# y >= 2
y1 = (x*0) + 2
# 2y <= 25 - x
y2 = (25-x)/2.0
# 4y >= 2x - 8
y3 = (2*x-8)/4.0
# y <= 2x - 5
y4 = 2 * x -5
# Make plot
plt.plot(x, 2*np.ones_like(y1))
plt.plot(x, y2, label=r'$2y\leq25-x$')
plt.plot(x, y3, label=r'$4y\geq 2x - 8$')
plt.plot(x, y4, label=r'$y\leq 2x-5$')
plt.xlim(0,16)
plt.ylim(0,11)
plt.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.)
plt.xlabel(r'$x$')
plt.ylabel(r'$y$')
This is a vertex enumeration problem. You can use the function lineqs which visualizes the system of inequalities A x >= b for any number of lines. The function will also display the vertices on which the graph was plotted.
The last 2 lines mean that x,y >=0
from intvalpy import lineqs
import numpy as np
A = -np.array([[5, 10],
[-1, 0],
[0, -1]])
b = -np.array([10, 0, 0])
lineqs(A, b, title='Solution', color='gray', alpha=0.5, s=10, size=(15,15), save=False, show=True)
Visual Solution Link
EDIT: I already made significant progress. My current question is written after my last edit below and can be answered without the context.
I currently follow Andrew Ng's Machine Learning Course on Coursera and tried to implement logistic regression today.
Notation:
X is a (m x n)-matrix with vectors of input variables as rows (m training samples of n-1 variables, the entries of the first column are equal to 1 everywhere to represent a constant).
y is the corresponding vector of expected output samples (column vector with m entries equal to 0 or 1)
theta is the vector of model coefficients (row vector with n entries)
For an input row vector x the model will predict the probability sigmoid(x * theta.T) for a positive outcome.
This is my Python3/numpy implementation:
import numpy as np
def sigmoid(x):
return 1 / (1 + np.exp(-x))
vec_sigmoid = np.vectorize(sigmoid)
def logistic_cost(X, y, theta):
summands = np.multiply(y, np.log(vec_sigmoid(X*theta.T))) + np.multiply(1 - y, np.log(1 - vec_sigmoid(X*theta.T)))
return - np.sum(summands) / len(y)
def gradient_descent(X, y, learning_rate, num_iterations):
num_parameters = X.shape[1] # dim theta
theta = np.matrix([0.0 for i in range(num_parameters)]) # init theta
cost = [0.0 for i in range(num_iterations)]
for it in range(num_iterations):
error = np.repeat(vec_sigmoid(X * theta.T) - y, num_parameters, axis=1)
error_derivative = np.sum(np.multiply(error, X), axis=0)
theta = theta - (learning_rate / len(y)) * error_derivative
cost[it] = logistic_cost(X, y, theta)
return theta, cost
This implementation seems to work fine, but I encountered a problem when calculating the logistic-cost. At some point the gradient descent algorithm converges to a pretty good fitting theta and the following happens:
For some input row X_i with expected outcome 1 X * theta.T will become positive with a good margin (for example 23.207). This will lead to sigmoid(X_i * theta) to become exactly 1.0000 (this is because of lost precision I think). This is a good prediction (since the expected outcome is equal to 1), but this breaks the calculation of the logistic cost, since np.log(1 - vec_sigmoid(X*theta.T)) will evaluate to NaN. This shouldn't be a problem, since the term is multiplied with 1 - y = 0, but once a value of NaN occurs, the whole calculation is broken (0 * NaN = NaN).
How should I handle this in the vectorized implementation, since np.multiply(1 - y, np.log(1 - vec_sigmoid(X*theta.T))) is calculated in every row of X (not only where y = 0)?
Example input:
X = np.matrix([[1. , 0. , 0. ],
[1. , 1. , 0. ],
[1. , 0. , 1. ],
[1. , 0.5, 0.3],
[1. , 1. , 0.2]])
y = np.matrix([[0],
[1],
[1],
[0],
[1]])
Then theta, _ = gradient_descent(X, y, 10000, 10000) (yes, in this case we can set the learning rate this large) will set theta as:
theta = np.matrix([[-3000.04008972, 3499.97995514, 4099.98797308]])
This will lead to vec_sigmoid(X * theta.T) to be the really good prediction of:
np.matrix([[0.00000000e+00], # 0
[1.00000000e+00], # 1
[1.00000000e+00], # 1
[1.95334953e-09], # nearly zero
[1.00000000e+00]]) # 1
but logistic_cost(X, y, theta) evaluates to NaN.
EDIT:
I came up with the following solution. I just replaced the logistic_cost function with:
def new_logistic_cost(X, y, theta):
term1 = vec_sigmoid(X*theta.T)
term1[y == 0] = 1
term2 = 1 - vec_sigmoid(X*theta.T)
term2[y == 1] = 1
summands = np.multiply(y, np.log(term1)) + np.multiply(1 - y, np.log(term2))
return - np.sum(summands) / len(y)
By using the mask I just calculate log(1) at the places at which the result will be multiplied with zero anyway. Now log(0) will only happen in wrong implementations of gradient descent.
Open questions: How can I make this solution more clean? Is it possible to achieve a similar effect in a cleaner way?
If you don't mind using SciPy, you could import expit and xlog1py from scipy.special:
from scipy.special import expit, xlog1py
and replace the expression
np.multiply(1 - y, np.log(1 - vec_sigmoid(X*theta.T)))
with
xlog1py(1 - y, -expit(X*theta.T))
I know it is an old question but I ran into the same problem, and maybe it can help others in the future, I actually solved it by implementing normalization on the data before appending X0.
def normalize_data(X):
mean = np.mean(X, axis=0)
std = np.std(X, axis=0)
return (X-mean) / std
After this all worked well!
I am working on a pedestrian tracking algorithm using Python3 & OpenCV.
We can use SIFT keypoints as an identifier of a pedestrian silhouette on a frame and then perform brute force matching between two sets of SIFT keypoints (i.e. between one frame and the next one) to find the pedestrian in the next frame.
To visualize this on the sequence of frames, we can draw a bounding rectangle delimiting the pedestrian. This is what it looks like :
The main problem is about characterizing the motion of the pedestrian using the keypoints. The idea here is to find an affine transform (that is translation in x & y, rotation & scaling) using the coordinates of the keypoints on 2 successives frames. Ideally, this affine transform somehow corresponds to the motion of the pedestrian. To track this pedestrian, we would then just have to apply the same affine transform on the bounding rectangle coordinates.
That last part doesn’t work well. The rectangle consistently shrinks over several frames to inevitably disappear or drifts away from the pedestrian, as you see below or on the previous image :
To specify, we characterize the bounding rectangle with 2 extreme points :
There are some built-in cv2 functions that can apply an affine transform to an image, like cv2.warpAffine(), but I want to apply it only to the bounding rectangle coordinates (i.e 2 points or 1 point + width & height).
To find the affine transform between the 2 sets of keypoints, I’ve written my own function (I can post the code if it helps), but I’ve observed similar results when using cv2.getAffineTransform() for instance.
Do you know how to properly apply an affine transform to this bounding rectangle ?
EDIT : here’s some explanation & code for better context :
The pedestrian detection is done with the pre-trained SVM classifier available in openCV : hog.setSVMDetector(cv2.HOGDescriptor_getDefaultPeopleDetector()) & hog.detectMultiScale()
Once a first pedestrian is detected, the SVM returns the coordinates of the associated bounding rectangle (xA, yA, w, h) (we stop using the SVM after the 1st detection as it is quite slow, and we are focusing on one pedestrian for now)
We select the corresponding region of the current frame, with image[yA: yA+h, xA: xA+w] and search for SURF keypoints within with surf.detectAndCompute()
This returns the keypoints & their associated descriptors (an array of 64 characteristics for each keypoint)
We perform brute force matching, based on the L2-norm between the descriptors and the distance in pixels between the keypoints to construct pairs of keypoints between the current frame & the previous one. The code for this function is pretty long, but should be similar to cv2.BFMatcher(cv2.NORM_L2, crossCheck=True)
Once we have the matched pairs of keypoints, we can use them to find the affine transform with this function :
previousKpts = previousKpts[:5] # select 4 best matches
currentKpts = currentKpts[:5]
# build A matrix of shape [2 * Nb of keypoints, 4]
A = np.ndarray(((2 * len(previousKpts), 4)))
for idx, keypoint in enumerate(previousKpts):
# Keypoint.pt = (x-coord, y-coord)
A[2 * idx, :] = [keypoint.pt[0], -keypoint.pt[1], 1, 0]
A[2 * idx + 1, :] = [keypoint.pt[1], keypoint.pt[0], 0, 1]
# build b matrix of shape [2 * Nb of keypoints, 1]
b = np.ndarray((2 * len(previousKpts), 1))
for idx, keypoint in enumerate(currentKpts):
b[2 * idx, :] = keypoint.pt[0]
b[2 * idx + 1, :] = keypoint.pt[1]
# convert the numpy.ndarrays to matrix :
A = np.matrix(A)
b = np.matrix(b)
# solution of the form x = [x1, x2, x3, x4]' = ((A' * A)^-1) * A' * b
x = np.linalg.inv(A.T * A) * A.T * b
theta = math.atan2(x[1, 0], x[0, 0]) # outputs rotation angle in [-pi, pi]
alpha = math.sqrt(x[0, 0] ** 2 + x[1, 0] ** 2) # scaling parameter
bx = x[2, 0] # translation along x-axis
by = x[3, 0] # translation along y-axis
return theta, alpha, bx, by
We then just have to apply the same affine transform to the corner points of the bounding rectangle :
# define the 4 bounding points using xA, yA
xB = xA + w
yB = yA + h
rect_pts = np.array([[[xA, yA]], [[xB, yA]], [[xA, yB]], [[xB, yB]]], dtype=np.float32)
# warp the affine transform into a full perspective transform
affine_warp = np.array([[alpha*np.cos(theta), -alpha*np.sin(theta), tx],
[alpha*np.sin(theta), alpha*np.cos(theta), ty],
[0, 0, 1]], dtype=np.float32)
# apply affine transform
rect_pts = cv2.perspectiveTransform(rect_pts, affine_warp)
xA = rect_pts[0, 0, 0]
yA = rect_pts[0, 0, 1]
xB = rect_pts[3, 0, 0]
yB = rect_pts[3, 0, 1]
return xA, yA, xB, yB
Save the updated rectangle coordinates (xA, yA, xB, yB), all current keypoints & descriptors, and iterate over the next frame : select image[yA: yB, xA: xA] using (xA, yA, xB, yB) we previously saved, get SURF keypoints etc.
As Micka suggested, cv2.perspectiveTransform() is an easy way to accomplish this. You'll just need to turn your affine warp into a full perspective transform (homography) by adding a third row at the bottom with the values [0, 0, 1]. For example, let's put a box with w, h = 100, 200 at the point (10, 20) and then use an affine transformation to shift the points so that the box is moved to (0, 0) (i.e. shift 10 pixels to the left and 20 pixels up):
>>> xA, yA, w, h = (10, 20, 100, 200)
>>> xB, yB = xA + w, yA + h
>>> rect_pts = np.array([[[xA, yA]], [[xB, yA]], [[xA, yB]], [[xB, yB]]], dtype=np.float32)
>>> affine_warp = np.array([[1, 0, -10], [0, 1, -20], [0, 0, 1]], dtype=np.float32)
>>> cv2.perspectiveTransform(rect_pts, affine_warp)
array([[[ 0., 0.]],
[[ 100., 0.]],
[[ 0., 200.]],
[[ 100., 200.]]], dtype=float32)
So that works perfectly as expected. You could also just simply transform the points yourself with matrix multiplication:
>>> rect_pts.dot(affine_warp[:, :2]) + affine_warp[:, 2]
array([[[ 0., 0.]],
[[ 100., 0.]],
[[ 0., 200.]],
[[ 100., 200.]]], dtype=float32)