def train(self, train_ds):
face_count = 0
nonface_count = 0
image_count = len(train_ds)
# face and nonface image count
for x in range(image_count):
label = train_ds[x][1]
if label == 1: # Face
face_count = face_count + 1
else:
nonface_count = nonface_count + 1
# loading integral images of train_ds
with open("train_integral_ds.pkl", 'rb') as f:
train_ds_integral = pickle.load(f)
print(
"Integral images are loaded from pickle file for training the model"
)
w = np.zeros(image_count) # sample_Weight
for x in range(image_count):
# Initial weight of every image (sample weight)
if label == 1: # Face
w[x] = 1.0 / (2 * face_count)
else:
w[x] = 1.0 / (2 * nonface_count)
# feature generation
f = Feature(train_ds[0][0].shape)
features = f.creating_all_features()
# load features value (X) and classification (y) from pickle file
# saves a lot of time while tuning in training phase
with open("features_value_1.pkl", 'rb') as f:
a = pickle.load(f)
with open("features_value_2.pkl", 'rb') as f:
b = pickle.load(f)
X = np.concatenate((a, b), axis=0)
with open("y.pkl", 'rb') as f:
y = pickle.load(f)
# Selecting weak classifiers
for wc_i in range(self.total_wc):
w = w / np.linalg.norm(w)
trainedWeakClassifier = self.weakClassifier_training(
X, y, features, w, face_count, nonface_count)
bc = None
bacc = None
be = float('inf')
for twc in trainedWeakClassifier:
e = 0
acc = []
for integral_image, ww in zip(train_ds_integral, w):
predicted_label = twc.classify(integral_image[0])
true_label = integral_image[1]
acc.append(abs(predicted_label - true_label))
e += ww * abs(predicted_label - true_label)
e = e / len(train_ds_integral)
if e < be:
bc = twc
be = e
bacc = acc
beta = be / (1.0 - be)
for i in range(len(bacc)):
w[i] = w[i] * (beta**(1 - bacc[i]))
alpha = math.log(1.0 / beta)
self.classifiersWeights.append(alpha)
self.classifiers.append(bc)
from rectangleArea import Rectangle
from feature import Feature
with open("train_ds.pkl", 'rb') as f:
train_ds = pickle.load(f)
with open("train_integral_ds.pkl", 'rb') as f:
train_ds_integral = pickle.load(f)
# creating all features for a certain window shape
s = time.time()
f = Feature(train_ds[0][0].shape)
features = f.creating_all_features()
e = time.time()
print("Feature creation is done")
print("Time for creating all features :")
print(e-s)
start = time.time()
X1, y1 = f.features_value(train_ds_integral)
end = time.time()
print("Feature value is calculated for all training images")
