def read_data(datasetdir, lv):
alcon = AlconUtils(datasetdir)
# アノテーションの読み込み
fn = "target_lv" + lv + ".csv"
alcon.load_annotations_target(fn)
fn = "groundtruth_lv" + lv + ".csv"
alcon.load_annotations_ground(fn)
# KNNモデルの作成
dataset = {}
for bb_id, target in alcon.targets.items():
img_filename = alcon.get_filename_char(bb_id)
code = alcon.ground_truth[bb_id][0]
if code not in dataset:
dataset[code] = []
#if len(dataset[code]) == 100:
# continue
img = cv2.imread(img_filename)
feature = MyAlgorithm.feature_extraction(img)
dataset[code].append(feature)
labels = []
data = []
classes = sorted(dataset.keys())
for label, values in dataset.items():
labels += [classes.index(label)] * len(values)
data += values
data = np.asarray(data, dtype=np.float)
labels = np.asarray(labels, dtype=np.int)
return data, labels, classes
def main(datasetdir,lv):
input_shape = (64, 64, 3)
# 初期化
alcon = AlconUtils(datasetdir)
# アノテーションの読み込み
fn = "target_lv" + lv + "_samp_0.9.csv"
alcon.load_annotations_target(fn)
fn = "groundtruth_lv" + lv + "_samp_0.9.csv"
alcon.load_annotations_ground(fn)
# CNNモデルの作成
dataset = {}
for bb_id, target in alcon.targets.items():
img_filename = alcon.get_filename_char( bb_id )
code = alcon.ground_truth[bb_id][0]
if code not in dataset:
dataset[code] = []
img = cv2.imread( img_filename )
img = cv2.resize(img, (input_shape[0], input_shape[1]))
img = np.asarray(img, dtype=np.float) / np.max(img)
dataset[code].append(img)
labels = []
data = []
classes = sorted(dataset.keys())
for label, values in dataset.items():
labels += [classes.index(label)] * len(values)
data += values
data = np.asarray(data, dtype=np.float)
labels = np.asarray(labels, dtype=np.int)
labels_c = keras.utils.to_categorical(labels)
# ネットワークの学習
model = Sequential()
model.add(Conv2D(64, kernel_size=(3, 3), activation='relu', input_shape=input_shape))
model.add(MaxPooling2D(pool_size=(3, 3)))
model.add(Conv2D(128, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(3, 3)))
model.add(Flatten())
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(len(classes), activation='softmax'))
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Adam(),
metrics=['accuracy'])
model.fit(data, labels_c, batch_size=100, epochs=100, verbose=1)
outputfile = os.path.join(datasetdir, "model.h5")
classfile = os.path.join(datasetdir, "classes.pkl")
model.save(outputfile)
joblib.dump(classes, classfile)
def main(datasetdir,lv):
# 初期化
alcon = AlconUtils(datasetdir)
# アノテーションの読み込み
fn = "target_lv" + lv + "_samp_5.csv"
alcon.load_annotations_target(fn)
fn = "groundtruth_lv" + lv + "_samp_5.csv"
alcon.load_annotations_ground(fn)
# KNNモデルの作成
dataset = {}
for bb_id, target in alcon.targets.items():
img_filename = alcon.get_filename_char( bb_id )
code = alcon.ground_truth[bb_id][0]
if code not in dataset:
dataset[code] = []
# if len(dataset[code]) == 5:
# continue
img = cv2.imread( img_filename )
feature = MyAlgorithm.feature_extraction(img)
dataset[code].append(feature)
labels = []
data = []
classes = sorted(dataset.keys())
for label, values in dataset.items():
labels += [classes.index(label)] * len(values)
data += values
data = np.asarray(data, dtype=np.float)
labels = np.asarray(labels, dtype=np.int)
classifier = KNeighborsClassifier() #編集
classifier.fit(data, labels)
outputfile = "./model.pkl"
joblib.dump((classes, classifier), outputfile)
def main(datasetdir, lv):
config = tensorflow.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.95
backend.tensorflow_backend.set_session(tensorflow.Session(config=config))
# 初期化
alcon = AlconUtils(datasetdir)
# アノテーションの読み込み
alcon.load_annotations_target("target_lv" + lv + ".csv")
alcon.load_annotations_ground("groundtruth_lv" + lv + ".csv")
dataset = {}
for bb_id, target in alcon.targets.items():
code = alcon.ground_truth[bb_id][0]
if code not in dataset:
dataset[code] = []
# if len(dataset[code]) == 10:
# continue
img_filename = alcon.get_filename_char(bb_id)
img = cv2.imread(img_filename)
feature = MyAlgorithm.feature_extraction(img)
dataset[code].append(feature)
labels = []
data = []
classes = sorted(dataset.keys())
for label, values in dataset.items():
labels += [classes.index(label)] * len(values)
data += values
batch_size = 128
num_classes = 46
epochs = 12
img_rows, img_cols = 32, 32
channel = 1
y_train = keras.utils.to_categorical(labels, num_classes)
classifier = Sequential()
classifier.add(
keras.layers.normalization.BatchNormalization(input_shape=(img_rows,
img_cols,
channel)))
classifier.add(Conv2D(32, (3, 3), activation='relu'))
classifier.add(Conv2D(64, (3, 3), activation='relu'))
classifier.add(MaxPooling2D(pool_size=(4, 4)))
classifier.add(Conv2D(64, (3, 3), activation='relu', padding='same'))
classifier.add(Conv2D(64, (3, 3), activation='relu', padding='same'))
classifier.add(Dropout(0.5))
classifier.add(Flatten())
classifier.add(Dense(128, activation='relu'))
classifier.add(Dropout(0.5))
classifier.add(Dense(num_classes, activation='softmax'))
classifier.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Nadam(),
metrics=['accuracy'])
classifier.summary()
numpy_data = numpy.asarray(data, dtype=numpy.float)
float_data = numpy_data.reshape(numpy_data.shape[0],img_rows,img_cols, channel )\
.astype('float32')
numpy_labels = numpy.asarray(labels, dtype=numpy.int)
y_train = keras.utils.to_categorical(numpy_labels, num_classes)
classifier.fit(float_data,
y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=None)
outputfile = "./model.pkl"
outputfile2 = "./model2.pkl"
joblib.dump(classes, outputfile)
classifier.save(outputfile2)
def main(datasetdir,lv):
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 1.0
K.tensorflow_backend.set_session(tf.Session(config=config))
# 初期化
alcon = AlconUtils(datasetdir)
# アノテーションの読み込み
fn = "target_lv" + lv + ".csv"
alcon.load_annotations_target(fn)
fn = "groundtruth_lv" + lv + ".csv"
alcon.load_annotations_ground(fn)
# KNNモデルの作成
dataset = {}
print(len(alcon.targets.items()))
for bb_id, target in alcon.targets.items():
img_filename = alcon.get_filename_char( bb_id )
code = alcon.ground_truth[bb_id][0]
if code not in dataset:
dataset[code] = []
if len(dataset[code]) == 100:
continue
img = cv2.imread( img_filename )
feature = MyAlgorithm.feature_extraction(img)
dataset[code].append(feature)
labels = []
data = []
classes = sorted(dataset.keys())
print(len(dataset.items()))
for label, values in dataset.items():
labels += [classes.index(label)] * len(values)
data += values
data = np.asarray(data, dtype=np.float)
labels = np.asarray(labels, dtype=np.int)
print(data.shape)
print(labels.shape)
# classifier = KNeighborsClassifier()
# classifier = RandomForestClassifier()
# classifier.fit(data, labels)
batch_size = 128
num_classes = 46
epochs = 12
img_rows, img_cols = 32, 32
# (x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = data
y_train = labels
x_train = x_train.reshape(x_train.shape[0],img_rows,img_cols,3)
# x_test = x_test.reshape(x_test.shape[0],img_rows,img_cols,1)
input_shape = (img_rows, img_cols, 3)
x_train = x_train.astype('float32')
# x_test = x_test.astype('float32')
# x_test /= 255
y_train = keras.utils.to_categorical(y_train, num_classes)
# y_test = keras.utils.to_categorical(y_test, num_classes)
classifier = Sequential()
classifier.add(Conv2D(32, kernel_size=(3,3), activation='relu', input_shape=input_shape,padding='same'))
classifier.add(Conv2D(64,(3,3),activation='relu',padding='same'))
classifier.add(MaxPooling2D(pool_size=(2,2)))
classifier.add(Dropout(0.25))
classifier.add(Flatten())
classifier.add(Dense(128,activation='relu'))
classifier.add(Dropout(0.5))
classifier.add(Dense(num_classes, activation='softmax'))
classifier.compile(loss=keras.losses.categorical_crossentropy,optimizer=keras.optimizers.Adadelta(),metrics=['accuracy'])
classifier.fit(x_train,y_train,batch_size=batch_size,epochs=epochs,verbose=1,validation_data=None)
# classifier.fit(data,labels)
outputfile = "./model.pkl"
outputfile2 = "./model2.pkl"
joblib.dump(classes, outputfile)
classifier.save(outputfile2)
def main(datasetdir, lv, length):
pr = cProfile.Profile()
pr.enable()
config = tensorflow.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.93
keras.backend.tensorflow_backend.set_session(
tensorflow.Session(config=config))
# 初期化
alcon = AlconUtils(datasetdir)
# アノテーションの読み込み
alcon.load_annotations_target("target_lv1_samp_0.9.csv")
alcon.load_annotations_ground("groundtruth_lv1_samp_0.9.csv")
dataset = {}
for bb_id, target in alcon.targets.items():
code = alcon.ground_truth[bb_id][0]
if code not in dataset:
dataset[code] = []
if len(dataset[code]) == int(length):
continue
img_filename = alcon.get_filename_char(bb_id)
img = cv2.imread(img_filename)
feature = MyAlgorithm.feature_extraction(img)
dataset[code].append(feature)
labels = []
data = []
classes = sorted(dataset.keys())
for label, values in dataset.items():
labels += [classes.index(label)] * len(values)
data += values
num_classes = 46
input_shape = (32, 32, 1) # img_rows img_cols channel
classifier = keras.models.Sequential()
classifier.add(
keras.layers.normalization.BatchNormalization(input_shape=input_shape))
classifier.add(Conv2D(32, kernel_size=(3, 3), activation='relu')) # 30*30
classifier.add(Conv2D(64, (3, 3), activation='relu')) # 28*28
classifier.add(Dropout(0.5))
classifier.add(MaxPooling2D(pool_size=(4, 4))) # 7*7
classifier.add(Flatten())
classifier.add(Dense(128, activation='relu'))
classifier.add(Dropout(0.5))
classifier.add(Dense(num_classes, activation='softmax'))
classifier.compile(loss=keras.losses.categorical_crossentropy,
optimizer=keras.optimizers.Nadam(),
metrics=['accuracy'])
classifier.summary()
x_data = np.asarray(data).reshape(len(data), *input_shape)
y_train = keras.utils.to_categorical(labels, num_classes)
classifier.fit(x_data, y_train, batch_size=84, epochs=12)
joblib.dump(classes, "./model.pkl")
classifier.save("./model2.pkl")
pr.disable()
pstats.Stats(pr).sort_stats('tottime').print_stats(5)
def load_data_target(self):
#if os.path.exists(self.dump_name + '1'):
# self.load_dataset()
if self.target is None:
self.target = []
self.data = []
self.hash_target = []
# 初期化
alcon = AlconUtils(self.data_dir_path)
# アノテーションの読み込み
fn = "target_lv" + self.dump_name + ".csv"
alcon.load_annotations_target(fn)
fn = "groundtruth_lv" + self.dump_name + ".csv"
alcon.load_annotations_ground(fn)
for bb_id, target in alcon.targets.items():
img_filename = alcon.get_filename_char(bb_id)
#print(img_filename)
code = alcon.ground_truth[bb_id]
# Load an color image in grayscale
img = cv2.imread(img_filename, 0)
#print(shape(img))
height, width = img.shape
WHITE = [255, 255, 255]
if height > width:
img = cv2.copyMakeBorder(img,
0,
0, (height - width) // 2,
int(np.ceil(
(height - width) / 2)),
cv2.BORDER_CONSTANT,
value=WHITE)
else:
img = cv2.copyMakeBorder(img, (width - height) // 2,
int(np.ceil(
(width - height) / 2)),
0,
0,
cv2.BORDER_CONSTANT,
value=WHITE)
#print(img.shape)
img.resize(self.image_size, self.image_size, 1)
#img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
self.data.append(img)
# determine index of target of this data
index_target = -1
try:
index_target = self.hash_target.index(code)
except ValueError:
self.hash_target.append(code)
index_target = self.hash_target.index(code)
self.target.append(index_target)
self.data = np.array(self.data, np.float32)
self.target = np.array(self.target, np.int32)
self.categories = len(self.hash_target)
print(self.data.shape)
print(self.target.shape)
print(self.categories)