def part_4_a_b():
pos = load_images_from_dir(POS_DIR)
neg = load_images_from_dir(NEG_DIR)
train_pos = pos[:35]
train_neg = neg[:]
images = train_pos + train_neg
labels = np.array(len(train_pos) * [1] + len(train_neg) * [-1])
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(train_pos, train_neg, integral_images)
VJ.createHaarFeatures()
VJ.train(4)
VJ.haarFeatures[VJ.classifiers[0].feature].preview(filename="ps6-4-b-1")
VJ.haarFeatures[VJ.classifiers[1].feature].preview(filename="ps6-4-b-2")
predictions = VJ.predict(images)
vj_accuracy = None
vj_accuracy = 100. * helper_func(labels, predictions_train) / len(labels)
print "Prediction accuracy on training: {0:.2f}%".format(vj_accuracy)
neg = load_images_from_dir(NEG2_DIR)
test_pos = pos[35:]
test_neg = neg[:35]
test_images = test_pos + test_neg
real_labels = np.array(len(test_pos) * [1] + len(test_neg) * [-1])
predictions = VJ.predict(test_images)
vj_accuracy = None
vj_accuracy = 100. * helper_func(real_labels, predictions_train) / len(real_labels)
print "Prediction accuracy on testing: {0:.2f}%".format(vj_accuracy)
def part_4_a_b():
pos = load_images_from_dir(POS_DIR)
neg = load_images_from_dir(NEG_DIR)
train_pos = pos[:35]
train_neg = neg[:]
images = train_pos + train_neg
labels = np.array(len(train_pos) * [1] + len(train_neg) * [-1])
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(train_pos, train_neg, integral_images)
VJ.createHaarFeatures()
VJ.train(5)
VJ.haarFeatures[VJ.classifiers[0].feature].preview(filename="ps6-4-b-1")
VJ.haarFeatures[VJ.classifiers[1].feature].preview(filename="ps6-4-b-2")
predictions = VJ.predict(images)
matching_indices = np.where(predictions == labels)[0]
vj_accuracy = matching_indices.shape[0] / labels.shape[0]
print("Prediction accuracy on training: {0:.2f}%".format(vj_accuracy))
neg = load_images_from_dir(NEG2_DIR)
test_pos = pos[35:]
test_neg = neg[:35]
test_images = test_pos + test_neg
real_labels = np.array(len(test_pos) * [1] + len(test_neg) * [-1])
predictions = VJ.predict(test_images)
matching_indices = np.where(predictions == real_labels)[0]
vj_accuracy = matching_indices.shape[0] / real_labels.shape[0]
print("Prediction accuracy on testing: {0:.2f}%".format(vj_accuracy))
def part_4_a_b():
pos = load_images_from_dir(POS_DIR)
neg = load_images_from_dir(NEG_DIR)
train_pos = pos[:35]
train_neg = neg[:]
images = train_pos + train_neg
labels = np.array(len(train_pos) * [1] + len(train_neg) * [-1])
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(train_pos, train_neg, integral_images)
VJ.createHaarFeatures()
VJ.train(5)
VJ.haarFeatures[VJ.classifiers[0].feature].preview(filename="ps6-4-b-1")
VJ.haarFeatures[VJ.classifiers[1].feature].preview(filename="ps6-4-b-2")
predictions = VJ.predict(images)
vj_accuracy = None
good = 0
bad = 0
for i in range(0, len(predictions)):
if(labels[i] == predictions[i]):
good += 1
else:
bad += 1
vj_accuracy = 100* float(good) / (good + bad)
print "Prediction accuracy on training: {0:.2f}%".format(vj_accuracy)
neg = load_images_from_dir(NEG2_DIR)
test_pos = pos[35:]
test_neg = neg[:35]
test_images = test_pos + test_neg
real_labels = np.array(len(test_pos) * [1] + len(test_neg) * [-1])
predictions = VJ.predict(test_images)
vj_accuracy = None
good = 0
bad = 0
for i in range(0, len(predictions)):
if(real_labels[i] == predictions[i]):
good += 1
else:
bad += 1
vj_accuracy = 100* float(good) / (good + bad)
print "Prediction accuracy on testing: {0:.2f}%".format(vj_accuracy)
def part_4_a_b():
pos = load_images_from_dir(POS_DIR)
neg = load_images_from_dir(NEG_DIR)
train_pos = pos[:35]
train_neg = neg[:]
images = train_pos + train_neg
labels = np.array(len(train_pos) * [1] + len(train_neg) * [-1])
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(train_pos, train_neg, integral_images)
VJ.createHaarFeatures()
VJ.train(4)
#
# print VJ.haarFeatures[VJ.classifiers[0].feature]
# print VJ.haarFeatures[VJ.classifiers[1].feature]
VJ.haarFeatures[VJ.classifiers[0].feature].preview(filename="ps6-4-b-1")
VJ.haarFeatures[VJ.classifiers[1].feature].preview(filename="ps6-4-b-2")
predictions = VJ.predict(images)
matches = [labels[i] == predictions[i] for i in range(0, len(predictions))]
corr = np.sum(np.array([int(x) for x in matches]))
vj_accuracy = corr / float(len(labels)) * 100
print "Prediction accuracy on training: {0:.2f}%".format(vj_accuracy)
neg = load_images_from_dir(NEG2_DIR)
test_pos = pos[35:]
test_neg = neg[:35]
test_images = test_pos + test_neg
real_labels = np.array(len(test_pos) * [1] + len(test_neg) * [-1])
predictions = VJ.predict(test_images)
# print real_labels
# print predictions
matches = [
real_labels[i] == predictions[i] for i in range(0, len(predictions))
]
matches = [int(x) for x in matches]
# print matches
corr = np.sum(np.array([int(x) for x in matches]))
print corr
vj_accuracy = corr / float(len(real_labels)) * 100
print "Prediction accuracy on testing: {0:.2f}%".format(vj_accuracy)
def part_4_c():
pos = load_images_from_dir(POS_DIR)[:20]
neg = load_images_from_dir(NEG_DIR)
images = pos + neg
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(pos, neg, integral_images)
VJ.createHaarFeatures()
VJ.train(4)
image = cv2.imread(os.path.join(INPUT_DIR, "man.jpeg"), -1)
image = cv2.resize(image, (120, 60))
VJ.faceDetection(image, filename="ps4-4-c-1")
def part_4_c():
pos, neg = generate_examples()
images = pos + neg
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(pos, neg, integral_images)
VJ.createHaarFeatures()
VJ.train(1)
image = cv2.imread(os.path.join(INPUT_DIR, "man.jpeg"), -1)
image = cv2.resize(image, (120, 60))
VJ.faceDetection(image, filename="ps6-4-c-1")
def part_4_a_b():
pos = load_images_from_dir(POS_DIR)
neg = load_images_from_dir(NEG_DIR)
train_pos = pos[:35]
train_neg = neg[:]
images = train_pos + train_neg
labels = np.array(len(train_pos) * [1] + len(train_neg) * [-1])
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(train_pos, train_neg, integral_images)
VJ.createHaarFeatures()
VJ.train(5)
VJ.haarFeatures[VJ.classifiers[0].feature].preview(filename="ps6-4-b-1")
VJ.haarFeatures[VJ.classifiers[1].feature].preview(filename="ps6-4-b-2")
predictions = VJ.predict(images)
vj_accuracy = None
correct = 0.0
for i in range(len(labels)):
if predictions[i] == labels[i]:
correct += 1.0
vj_accuracy = (correct / len(labels)) * 100
print "Prediction accuracy on training: {0:.2f}%".format(vj_accuracy)
neg = load_images_from_dir(NEG2_DIR)
test_pos = pos[35:]
test_neg = neg[:35]
test_images = test_pos + test_neg
real_labels = np.array(len(test_pos) * [1] + len(test_neg) * [-1])
predictions = VJ.predict(test_images)
vj_accuracy = None
correct = 0.0
for i in range(len(real_labels)):
if predictions[i] == real_labels[i]:
correct += 1.0
vj_accuracy = (correct / len(real_labels)) * 100
print "Prediction accuracy on testing: {0:.2f}%".format(vj_accuracy)
def part_4_c():
pos = load_images_from_dir(POS_DIR)[:20]
neg = load_images_from_dir(NEG_DIR)
images = pos + neg
print('test shape: ', images[0].shape)
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(pos, neg, integral_images)
VJ.createHaarFeatures()
VJ.train(4)
image = cv2.imread(os.path.join(INPUT_DIR, "man2.jpeg"), -1)
resized_image = image.copy()
resized_image = cv2.resize(resized_image, (120, 60))
VJ.faceDetection(resized_image, filename="ps4-4-c-1", original_image=image)
def part_4_a_b():
pos = load_images_from_dir(POS_DIR)
neg = load_images_from_dir(NEG_DIR)
train_pos = pos[:35]
train_neg = neg[:]
images = train_pos + train_neg
labels = np.array(len(train_pos) * [1] + len(train_neg) * [-1])
integral_images = ps6.convert_images_to_integral_images(images)
VJ = ps6.ViolaJones(train_pos, train_neg, integral_images)
VJ.createHaarFeatures()
VJ.train(5)
VJ.haarFeatures[VJ.classifiers[0].feature].preview(filename="ps6-4-b-1")
VJ.haarFeatures[VJ.classifiers[1].feature].preview(filename="ps6-4-b-2")
predictions = VJ.predict(images)
temp_vj = np.zeros_like(predictions)
temp_vj[predictions == labels] = 1
vj_accuracy = 100 * float(np.sum(temp_vj) / len(temp_vj)) #None
# raise NotImplementedError
print("Prediction accuracy on training: {0:.2f}%".format(vj_accuracy))
neg = load_images_from_dir(NEG2_DIR)
test_pos = pos[35:]
test_neg = neg[:35]
test_images = test_pos + test_neg
real_labels = np.array(len(test_pos) * [1] + len(test_neg) * [-1])
predictions = VJ.predict(test_images)
temp_vj = np.zeros_like(predictions)
temp_vj[predictions == real_labels] = 1
vj_accuracy = 100 * float(np.sum(temp_vj) / len(real_labels)) #None
# raise NotImplementedError
print("Prediction accuracy on testing: {0:.2f}%".format(vj_accuracy))