def browse(images_path = './test', output_dir = './store', model_path = './outputs-0.ckpt/0-81600', n = 5, normal_size = 224, focal_degree = 0, method='quick', review = False, photo = 'raw', admin = False):
'''
designed for numerous images
it will output the result to the output_dir like:
0 : area img1(2) img2(box ,label)
1
2
3
...
'''
area=[] # matrix of 0 1
predicted_label_dict=[] # predictedlabel
predicted_logits_dict=[] #predicted logits
images_1=[]#1 images(crack)
logits_dict=[]#logits
launch=n*normal_size
if not os.path.exists(images_path):
raise ValueError('images_path is not exist.')
Images = []
images_path = os.path.join(images_path, '*.jpg')
count = 0
for image_file in glob.glob(images_path):
count += 1
if count % 100 == 0:
print('Load {} images.'.format(count))
image = cv2.imread(image_file)
Images.append(image)
Images = np.array(Images)
k=0
predicted_label_dict=[]
predicted_logits_dict=[]
with tf.Session() as sess:
ckpt_path = model_path
saver = tf.train.import_meta_graph(ckpt_path + '.meta')
saver.restore(sess, ckpt_path)
inputs = tf.get_default_graph().get_tensor_by_name('inputs:0')
classes = tf.get_default_graph().get_tensor_by_name('classes:0')
logits = tf.get_default_graph().get_tensor_by_name('logits:0')#logits
for i in range(len(Images)):
for j in range(n*n):
area.append(0)
image = cv2.resize(Images[i],(launch,launch),interpolation=cv2.INTER_CUBIC)
cv2.imwrite(output_dir+'/'+str(k)+'/image'+'.jpg',image)
images=utils2.pic_cut(image,n,normal_size=224)
for i in range(len(images)):
image_batch=[images[i]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
packed_images=utils2.pack_images_up(images,photo)
area,image,logits_dict,images_1=utils2.resort(area,packed_images,predicted_label_dict,predicted_logits_dict,image)
utils2.record(k,area,packed_images,image,output_dir)
area.clear()
predicted_label_dict.clear()
k+=1
def focal(image, model_path, focal_length, photo = 'raw', model = 'quick', admin = False):
'''
focue_detection:
auto for best cut search
inputs:image
model_path
focal_length
outputs:
area,image,logits_dict,images_1,stop
preference:
accuracy:more confidence
aomount:more amount
'''
raw_image=image
max_target=0
max_avg_score=0
#loss_=0
target=0
avg_score=0
score=0
stop=0
area=[] #
predicted_label_dict=[] #
predicted_logits_dict=[] #
logits_dict=[]
images_1=[]
normal_size=224
images_=[]
predicted_label_dict_=[]
predicted_logits_dict_=[]
if admin==True:
model=input('Input the model you prefer (you hava to select quick or slow):')
#preference = input('Input the preference (you hava to select accuray or amount):')
with tf.Session() as sess:
ckpt_path = model_path
saver = tf.train.import_meta_graph(ckpt_path + '.meta')
saver.restore(sess, ckpt_path)
inputs = tf.get_default_graph().get_tensor_by_name('inputs:0')
classes = tf.get_default_graph().get_tensor_by_name('classes:0')
logits = tf.get_default_graph().get_tensor_by_name('logits:0')#logits
if model=='quick':
while True :
for i in range(2,focal_length,1):
target=0
score=0
avg_score=0
predicted_label_dict.clear()
predicted_logits_dict.clear()
launch=i*normal_size #
image = cv2.resize(raw_image,(launch,launch),interpolation=cv2.INTER_CUBIC)
images=utils2.pic_cut(image,i,normal_size=224)
#predicted_label_dict,predicted_logits_dict=cnn_predict(images,model_path)
'''
if len(images)<50:
for o in range(len(images)):
image_batch=[images[o]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
else:
o=0
t=len(images)
while t>50:
t=t-50
o+=50
image_batch=[images[o:o+50]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
image_batch=[images[o:len(images)]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
'''
for o in range(len(images)):
image_batch=[images[o]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
for j in range(len(predicted_label_dict)):
if predicted_label_dict[j]==1:
target+=1
#print(predicted_logits_dict[j])
score=score+utils2.confidence_(predicted_logits_dict[j][0])
avg_score=score/target
print(target,avg_score,i)
if i==2 :
max_target=target
max_avg_score=avg_score
images_=images
predicted_label_dict_=predicted_label_dict
stop=2
else:
images_=images
predicted_label_dict_=predicted_label_dict
predicted_logits_dict_=predicted_logits_dict
stop=i
if utils2.gradient_function(max_target,max_avg_score,target,avg_score,preference='accuracy'):
break
else :
max_avg_score=avg_score
max_target=target
break
else:
for i in range(2,focal_length,1):
target=0
score=0
avg_score=0
predicted_label_dict.clear()
predicted_logits_dict.clear()
launch=i*normal_size #
image = cv2.resize(raw_image,(launch,launch),interpolation=cv2.INTER_CUBIC)
images=utils2.pic_cut(image,i,normal_size=224)
#predicted_label_dict,predicted_logits_dict=cnn_predict(images,model_path)
'''
if len(images)<50:
for o in range(len(images)):
image_batch=[images[o]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
else:
o=0
t=len(images)
while t>50:
t=t-50
o+=50
image_batch=[images[o:o+50]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
image_batch=[images[o:len(images)]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
'''
for o in range(len(images)):
image_batch=[images[o]]
predicted_label = sess.run(classes, feed_dict={inputs: image_batch})
predicted_label_dict.append(predicted_label)
predicted_logits = sess.run(logits, feed_dict={inputs: image_batch})
predicted_logits_dict.append(predicted_logits)
for j in range(len(predicted_label_dict)):
if predicted_label_dict[j]==1:
target+=1
#print(predicted_logits_dict[j])
score=score+utils2.confidence_(predicted_logits_dict[j][0])
avg_score=score/target
print(target,avg_score,i)
if i==2 :
max_target=target
max_avg_score=avg_score
images_=images
predicted_label_dict_=predicted_label_dict
stop=2
else:
images_=images
predicted_label_dict_=predicted_label_dict
predicted_logits_dict_=predicted_logits_dict
stop=i
if utils2.max_function(max_target,max_avg_score,target,avg_score,preference='accuracy'):
max_avg_score=avg_score
max_target=target
images_=utils2.pack_images_up(images_,photo)
for i in range(stop*stop):
area.append(0)
launch=stop*normal_size #
image = cv2.resize(raw_image,(launch,launch),interpolation=cv2.INTER_CUBIC)
for i in range(len(predicted_label_dict_)):
if predicted_label_dict_[i]==1:
area[i]=1;
logits_dict.append(predicted_logits_dict_[i])
images_1.append(images_[i])
y=normal_size*(int((i)/stop))
x=normal_size*((i)%stop)
cv2.rectangle(image,(x,y),(x+normal_size,y+normal_size),(0,255,0),3)
'''
area,image,logits_dict,images_1=utils.resort(area,packed_images,predicted_label_dict,predicted_logits_dict,image)
'''
return area,image,logits_dict,images_1,stop
def detection(image, model_path, n, normal_size = 224, focal_degree = 0, method = 'quick', review = False, photo = 'gradient', admin = False):
'''
inputs:image
model_path
number of cut_pic
normal_size
focal or not
method
review or not
gradient or not
administer or not
outputs:area(express crack of the image)
image:labeled image
images_:gradient target
logits_dict: logits of the
area:a map of the target in the image
'''
stop = 0
area = [] # matrix of 0 1
predicted_label_dict = [] # predictedlabel
predicted_logits_dict = [] #predicted logits
images_1 = []#1 images(crack)
images_0 = []#0 images(none)
logits_dict = []#logits
launch = n * normal_size
#fine tune
if focal_degree != 0:
#自动对焦模式
print('Focal starts ! It will take longer time...')
area, image, logits_dict, images_1, stop = focal(image, model_path, focal_degree, photo, admin)
return area,image,logits_dict,images_1,stop
else:
image = cv2.resize(image, (launch, launch), interpolation = cv2.INTER_CUBIC)
images = utils2.pic_cut(image, n, normal_size)
#images = utils2.simple_cut(image, m, n)
for i in range(n * n):
area.append(0)
#make the map
#标准模式
if method !='quick':
print('sorry,deep-method is still in f*****g testing' )
if method == 'quick':
predicted_label_dict,predicted_logits_dict = cnn_predict(images, model_path)#predict
packed_images = utils2.pack_images_up(images, photo)
area,image,logits_dict,images_1 = utils2.resort(area, packed_images, predicted_label_dict, predicted_logits_dict, image, normal_size)
else:
'''
deep:
muilt-random-detection
'''
raw_image=0
raw_image=image
predicted_label_dict,predicted_logits_dict=cnn_predict(images,model_path)#predict
#predict
packed_images=utils2.pack_images_up(images,photo)
#need reconsitution here
for i in range(len(images)):
if predicted_label_dict[i]==1:
area[i]=1
logits_dict.append(predicted_logits_dict[i])
images_1.append(packed_images[i])
y=normal_size*(int((i)/n))
x=normal_size*((i)%n)
cv2.rectangle(image,(x,y),(x+normal_size,y+normal_size),(0,255,0),3)
else:
images_0.append(images[i])
target,avg_confidence=utils2.confidence(predicted_label_dict,predicted_logits_dict)
print(target,avg_confidence)
patience = int(input('Input the patience (0 ~ 10 interger):'))
with tf.Session() as sess:
ckpt_path = model_path
saver = tf.train.import_meta_graph(ckpt_path + '.meta')
saver.restore(sess, ckpt_path)
inputs = tf.get_default_graph().get_tensor_by_name('inputs:0')
classes = tf.get_default_graph().get_tensor_by_name('classes:0')
logits = tf.get_default_graph().get_tensor_by_name('logits:0')#logits
for i in range(0,patience+n,1):
random_image,x,y=utils2.random_cut_(raw_image,normal_size)
#cv2.imshow('0',random_image)
label = sess.run(classes, feed_dict={inputs: [random_image]})
logits_ = sess.run(logits, feed_dict={inputs: [random_image]})
print(utils2.confidence_(logits_[0]))
if label==1:
if utils2.confidence_(logits_[0])>avg_confidence:
cv2.rectangle(image,(x,y),(x+normal_size,y+normal_size),(255,255,0),3)
images_1.append(utils2.pack_image_up_(random_image,photo))
logits_dict.append(logits_)
if review:
area,logits_dict,images_1_=review(images,model_path,area,logits_dict,images_1,images_0)
return area,image,logits_dict,images_1,stop