def edgeDetection(image):
image = imageresize.resize(image, height=500)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (5, 5), 0)
edged = cv2.Canny(gray, 100, 220)
# cv2.imshow("Image", image)
# cv2.imshow("Edged", edged)
# cv2.waitKey(0)
# cv2.destroyAllWindows()
return edged
def test_nofile(self):
""" 파일이 없는 경우 test """
error = imageresize.resize('nofile', './upload', [])
self.assertEqual(error[0], -100)
def test_split(self):
""" image resize test """
result = []
error = imageresize.resize('./test/1024_768.png', './upload', result)
self.assertEqual(error[0], 0)
def test_noimagefile(self):
""" 이미지 파일이 아닌 경우 test """
error = imageresize.resize('test.py', './upload', [])
self.assertEqual(error[0], -200)
def UseConvNet(image):
#define image path
image = str(image)
#resize the image to 277 by 277 so it fits our model
resize(image)
#Import Image
image = cv2.imread(image)
# cast image, which is now an array of intergers to an array of floats, intergers where giving a error
X = image.astype('float')
print("dtype is : ",X.dtype)
#preprosses image to have all values on a scale from 0-1
X = X * (1.0/255)
#define placeholders for model
x = tf.placeholder('float')
y = tf.placeholder('int64')
#
# Create model that the weights from teh save file will be aplleided to
def conv_net(x, n_classes, is_training):
x = tf.reshape(x, shape=[-1,227,227,1])
# Convolution Layer with 128 filters and a kernel size of 5
conv1 = tf.layers.conv2d(x, 128, 5, activation=tf.nn.relu)
# Max Pooling (down-sampling) with strides of 2 and kernel size of 2
conv1 = tf.layers.max_pooling2d(conv1, 10, 5,padding='same')
# Convolution Layer with 32 filters and a kernel size of 5
conv2 = tf.layers.conv2d(conv1, 128*2, 3, activation=tf.nn.relu)
# Max Pooling (down-sampling) with strides of 2 and kernel size of 2
conv2 = tf.layers.max_pooling2d(conv2, 20, 10,padding='same')
# Convolution Layer with 32 filters and a kernel size of 5
conv3 = tf.layers.conv2d(conv2, 128*2*2, 3, activation=tf.nn.relu)
# Max Pooling (down-sampling) with strides of 2 and kernel size of 2
conv3 = tf.layers.max_pooling2d(conv3, 20, 10,padding='same')
# Flatten the data to a 1-D vector for the fully connected layer
fc1 = tf.contrib.layers.flatten(conv3)
# Fully connected layer
fc1 = tf.layers.dense(fc1, 2048)
# Fully connected layer
fc1 = tf.layers.dense(fc1, 2048)
# Apply Dropout (if is_training is False, dropout is not applied)
#fc1 = tf.layers.dropout(fc1, rate=dropout, training=is_training)
# Output layer, class prediction
out = tf.layers.dense(fc1, n_classes)
# Because 'softmax_cross_entropy_with_logits' already apply softmax,
# I only apply softmax to testing network
#out = tf.nn.softmax(out)
return out
saver = tf.train.import_meta_graph('./model/model.ckpt.meta')
def use_neural_network(input_data):
prediction = conv_net(x, 2, is_training=False)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver.restore(sess,"./model/model.ckpt")
# pos: [1,0] , argmax: 0
# neg: [0,1] , argmax: 1
result = sess.run(tf.argmax(prediction.eval(feed_dict={x:input_data}),1))
print(result)
if result[0] < result[1]:
print('Positive:')
elif result[0] > result[1]:
print('Negative:')
elif result[0] == result[1]:
print("Can not tell")
else:
print("Ops somthing went wrong ¯/_(ツ)_/¯")
use_neural_network(X)