In [1]:
#%load ex4_objectdetection.py
In [2]:
%matplotlib inline
import matplotlib
matplotlib.rcParams['savefig.dpi'] = 2 * matplotlib.rcParams['savefig.dpi']
import numpy as np
from scipy.misc import imresize
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from matplotlib.patches import Rectangle
from sklearn_theano.feature_extraction import OverfeatLocalizer
from sklearn_theano.feature_extraction import OverfeatTransformer
Install sklearn-theano with this command:
pip install git+https://github.com/sklearn-theano/sklearn-theano
This example modified from sklearn-theano:
http://sklearn-theano.github.io/auto_examples/plot_localization_tutorial.html
In [3]:
def convert_points_to_box(points, color, alpha, window_adjust=0):
upper_left_point = (points[:, 0].min(), points[:, 1].min())
width = points[:, 0].max() - points[:, 0].min()
height = points[:, 1].max() - points[:, 1].min()
return Rectangle(upper_left_point, width, height, ec=color,
fc=color, alpha=alpha)
# Show the original image
f, axarr = plt.subplots(2, 2)
# Data needs to be float32 and 0 - 255
X = mpimg.imread("sloth.png").astype('float32')
# Shrink the image so that processing is faster
X = imresize(X, .4)
# Zero pad with half of input size on each size
pad_size = 230 // 2
XL = np.zeros((2 * pad_size + X.shape[0], 2 * pad_size + X.shape[1], 3),
dtype='float32')
XL[pad_size:pad_size + X.shape[0], pad_size:pad_size + X.shape[1], :] = X
X = XL
print("Read image")
print("Image size %sx%sx%s" % X.shape)
axarr[0, 0].imshow(X / 255.)
axarr[0, 0].axis('off')
# Show a single box
axarr[0, 1].imshow(X / 255.)
axarr[0, 1].axis('off')
r = Rectangle((0, 0), 231, 231, fc='yellow', ec='black', alpha=.8)
axarr[0, 1].add_patch(r)
# Show all the boxes being processed
axarr[1, 0].imshow(X / 255.)
axarr[1, 0].axis('off')
# Hard code box size to speed up processing
x_points = np.linspace(0, X.shape[1] - 231, 13)
y_points = np.linspace(0, X.shape[0] - 231, 10)
xx, yy = np.meshgrid(x_points, y_points)
for x, y in zip(xx.flat, yy.flat):
axarr[1, 0].add_patch(Rectangle((x, y), 231, 231, fc='yellow', ec='black',
alpha=.4))
print("Starting localization")
# Get all points with sloth in the top 5 labels
sloth_label = "three-toed sloth, ai, Bradypus tridactylus"
clf = OverfeatLocalizer(match_strings=[sloth_label])
sloth_points = clf.predict(X)[0]
axarr[1, 1].imshow(X / 255.)
axarr[1, 1].axis('off')
axarr[1, 1].autoscale(enable=False)
axarr[1, 1].scatter(sloth_points[:, 0], sloth_points[:, 1], color='orange',
s=50)
print("Localization complete!")
Original source: From Large Scale Image Categorization to Entry-Level Categories V. Ordonez, J. Deng, Y. Choi, A. Berg, T. Berg
In [4]:
from sklearn_theano.datasets import load_sample_image
from sklearn.mixture import GMM
def convert_gmm_to_box(gmm, color, alpha):
midpoint = gmm.means_
std = 3 * np.sqrt(clf.covars_)
width = std[:, 0]
height = std[:, 1]
upper_left_point = (midpoint[:, 0] - width // 2,
midpoint[:, 1] - height // 2)
return Rectangle(upper_left_point, width, height, ec=color,
fc=color, alpha=alpha)
X = load_sample_image("cat_and_dog.jpg")
dog_label = 'dog.n.01'
cat_label = 'cat.n.01'
print("Starting localization")
clf = OverfeatLocalizer(top_n=1,
match_strings=[dog_label, cat_label])
points = clf.predict(X)
print("Localization complete!")
dog_points = points[0]
cat_points = points[1]
plt.imshow(X)
ax = plt.gca()
ax.autoscale(enable=False)
clf = GMM()
clf.fit(dog_points)
dog_box = convert_gmm_to_box(clf, "darkred", .6)
clf.fit(cat_points)
cat_box = convert_gmm_to_box(clf, "steelblue", .6)
ax.add_patch(dog_box)
ax.add_patch(cat_box)
ax.get_xaxis().set_ticks([])
ax.get_yaxis().set_ticks([])
Out[4]:
In [4]: