def visualize():
print("Load images")
v_ims1 = common.load_images("data/vehicles/", color='RGB')
v_ims2 = common.load_images("data/OwnCollection/vehicles/", color='RGB')
nv_ims1 = common.load_images("data/non-vehicles/", color='RGB')
nv_ims2 = common.load_images("data/OwnCollection/non-vehicles/",
color='RGB')
ncols, nrows = (16, ) * 2
n = ncols * nrows // 4
ims = [
v_ims1[np.random.randint(v_ims1.shape[0], size=n)],
v_ims2[np.random.randint(v_ims2.shape[0], size=n)],
nv_ims1[np.random.randint(nv_ims1.shape[0], size=n)],
nv_ims2[np.random.randint(nv_ims2.shape[0], size=n)]
]
print("Load images done")
ims = np.concatenate(ims)
fig, axes = plt.subplots(ncols=ncols, nrows=nrows)
fig.canvas.set_window_title('Dataset images')
print("Visualize images {0}x{0}".format(ncols))
size = ims.shape[0]
for r, row_ax in enumerate(axes):
for c, ax in enumerate(row_ax):
i = r * ncols + c
ax.imshow(ims[i])
ax.get_xaxis().set_visible(False)
ax.get_yaxis().set_visible(False)
ax.set_xticklabels([])
ax.set_yticklabels([])
#ax.set_aspect('equal')
fig.subplots_adjust(wspace=0, hspace=0)
print("Show images")
fig.show()
def train(extract_features,
if_grayscale=False,
if_normalize_images=True,
svm_c_param=SVM_C_PARAM,
svm_gamma_param=SVM_GAMMA_PARAM,
**kwargs):
# Load images
(images, labels) = load_images(if_grayscale=if_grayscale,
if_normalize_images=if_normalize_images)
# Split images into train and test set
test_set_ratio = TEST_SET_RATIO
print('Splitting dataset to train & test set with test_set_ratio=' +
str(test_set_ratio) + '...')
(images_train, images_test, labels_train, labels_test) = \
model_selection.train_test_split(images, labels, test_size=test_set_ratio)
# Extract features from images
print('Extracting features for train & test set...')
(X_train, X_test) = extract_features(images_train, images_test, **kwargs)
y_train = labels_train
y_test = labels_test
# Visualize features
if IF_VISUALIZE_FEATURES:
from .visualize_features import visualize_features
visualize_features(X_train, y_train)
# Train SVM model with train set
gamma = svm_gamma_param if svm_gamma_param != 0 else 'auto'
print('Training SVM with c=' + str(svm_c_param) + ', gamma=' +
str(svm_gamma_param) + '...')
model = svm.SVC(max_iter=MAX_ITERATIONS,
C=svm_c_param,
gamma=gamma,
probability=True)
model.fit(X_train, y_train)
# Return the distance of the sample from all other hyperplane
# Higher value of the distance higher confidence
confidence_of_model = model.predict_proba(X_train)
confidence_example = np.amax(confidence_of_model, axis=1)
print(stats.describe(confidence_example))
#print(str(confidence_average))
#print(str(confidence_of_model))
# Test SVM model
y_train_predict = np.array(model.predict(X_train))
y_test_predict = np.array(model.predict(X_test))
(train_accuracy,
train_confusion_matrix) = get_accuracy(y_train_predict, y_train)
(test_accuracy,
test_confusion_matrix) = get_accuracy(y_test_predict, y_test)
print('Train accuracy: ' + str(train_accuracy))
print('Train confusion matrix:')
print(train_confusion_matrix)
print('Test accuracy: ' + str(test_accuracy))
print('Test confusion matrix:')
print(test_confusion_matrix)
return test_accuracy
def __init__(self, load=True,
vehicles=["data/vehicles/"],
nonvehicles=["data/non-vehicles"],
color='HSL'):
if load == True:
self.color = color
veh = common.load_images(*vehicles, color=color)
nonveh = common.load_images(*nonvehicles, color=color)
veh_lbl = np.array([1] * veh.shape[0])
nonveh_lbl = np.array([0] * nonveh.shape[0])
self.x_orig = np.concatenate([veh, nonveh])
self.y_orig = np.concatenate([veh_lbl, nonveh_lbl])
else:
self.x_orig = None
self.y_orig = None
self.x = None
self.colormap = color
def __init__(self, screen):
pygame.sprite.Sprite.__init__(self)
self.counter = 0
self.imgs = common.load_images('TN2.png', 'TE2.png',
'TS2.png', 'TW2.png',
'TNW2.png', 'TSW2.png',
'TSE2.png', 'TNE2.png')
self.masks = {}
for item in self.imgs:
self.masks[self.counter] = pygame.mask.from_surface(self.imgs[self.counter])
self.counter += 1
self.image = pygame.image.load('../resources/images/TN2.png').convert_alpha()
PlayerTank.__init__(self, screen)
def __init__(self, screen):
pygame.sprite.Sprite.__init__(self)
self.counter = 0
self.imgs = common.load_images('TN2.png', 'TE2.png', 'TS2.png',
'TW2.png', 'TNW2.png', 'TSW2.png',
'TSE2.png', 'TNE2.png')
self.masks = {}
for item in self.imgs:
self.masks[self.counter] = pygame.mask.from_surface(
self.imgs[self.counter])
self.counter += 1
self.image = pygame.image.load(
'../resources/images/TN2.png').convert_alpha()
PlayerTank.__init__(self, screen)
def train(extract_features, if_grayscale=False, if_normalize_images=True,
svm_c_param=SVM_C_PARAM, svm_gamma_param=SVM_GAMMA_PARAM, **kwargs):
# Load images
(images, labels) = load_images(if_grayscale=if_grayscale, if_normalize_images=if_normalize_images)
# Split images into train and test set
test_set_ratio = TEST_SET_RATIO
print('Splitting dataset to train & test set with test_set_ratio=' + str(test_set_ratio) + '...')
(images_train, images_test, labels_train, labels_test) = \
model_selection.train_test_split(images, labels, test_size=test_set_ratio)
# Extract features from images
print('Extracting features for train & test set...')
(X_train, X_test) = extract_features(images_train, images_test, **kwargs)
y_train = labels_train
y_test = labels_test
# Visualize features
if IF_VISUALIZE_FEATURES:
from .visualize_features import visualize_features
visualize_features(X_train, y_train)
# Train SVM model with train set
gamma = svm_gamma_param if svm_gamma_param != 0 else 'auto'
print('Training SVM with c=' + str(svm_c_param) + ', gamma=' + str(svm_gamma_param) + '...')
model = svm.SVC(max_iter=MAX_ITERATIONS, C=svm_c_param, gamma=gamma, probability=True)
model.fit(X_train, y_train)
# Return the distance of the sample from all other hyperplane
# Higher value of the distance higher confidence
confidence_of_model = model.predict_proba(X_train)
confidence_example = np.amax(confidence_of_model, axis=1)
print(stats.describe(confidence_example))
#print(str(confidence_average))
#print(str(confidence_of_model))
# Test SVM model
y_train_predict = np.array(model.predict(X_train))
y_test_predict = np.array(model.predict(X_test))
(train_accuracy, train_confusion_matrix) = get_accuracy(y_train_predict, y_train)
(test_accuracy, test_confusion_matrix) = get_accuracy(y_test_predict, y_test)
print('Train accuracy: ' + str(train_accuracy))
print('Train confusion matrix:')
print(train_confusion_matrix)
print('Test accuracy: ' + str(test_accuracy))
print('Test confusion matrix:')
print(test_confusion_matrix)
return test_accuracy
"""Face dataset
Attributes
----------
IMAGES : np.array
Data matrix for the faces dataset. 640x2500.
TEST_IMAGES : np.array
Data matrix for test images. 2x2500.
"""
from common import load_images
_IMAGE_NAMES = ["data/{idx}.png".format(idx=str(i + 1)) for i in range(640)]
IMAGES = load_images(_IMAGE_NAMES)
_TESTS = ["testcase/test_01.png", "testcase/test_02.png"]
TEST_IMAGES = load_images(_TESTS)
# -*- coding: utf-8 -*-
import numpy as np
import matplotlib.pyplot as plt
import common
import seaborn as sns
setname = 'DENIS'
size = (256, 256) if setname in ['SDO', 'SDO-Hard'] else None
fformat = '.jpg' if setname in ['CELEBa', 'INRIA'] else '.png'
training = common.load_images(f'../Clean/sets/{setname}-training', fformat)
testing = common.load_images(f'../Clean/sets/{setname}-testing', fformat)
#%%
mses = np.zeros(len(testing))
for i, x in enumerate(testing):
err = np.mean(np.mean((training - x)**2, 1), 1)
mses[i] = err.min()
#%%
plt.figure()
plt.title(f'{setname} : {np.mean(mses):0.2e}')
sns.distplot(mses)
plt.xlabel('Closest Error')
plt.ylabel('Frequency')
plt.grid()
plt.savefig(f'alignement_{setname}.eps')
np.savetxt(f"{setname}_align.csv", mses)