for i in range(len(data)):
temp = center-data[i,:]
temp_value = [email protected]
temp_dis[i] = np.sqrt(temp_value)
dis = np.max(temp_dis)
return dis
#plot the rectangle in the original image
def plot_rectangle(draw,area_x1,area_x2,area_y1,area_y2):
#draw = ImageDraw.Draw(im)
draw.line((area_y1, area_x1, area_y1, area_x2), fill="blue", width=10)
draw.line((area_y1, area_x2, area_y2, area_x2), fill="blue", width=10)
draw.line((area_y2, area_x2, area_y2, area_x1), fill="blue", width=10)
draw.line((area_y2, area_x1, area_y1, area_x1), fill="blue", width=10)
#im.show()
#%%load model and weight
model = cnn_model()
model.summary()
model.load_weights('./model_weight/2012images-areas-7.5-50epoch.h5')
#%%
windowsize_r = 30
windowsize_c = 30
raw_image_files = glob.glob("./New folder/*.jpg")
count = 0
for name in raw_image_files:
name_piece = name.split("\\")[1]
name_image = name_piece.split(".")[0]
path_image = './New folder/'+str(name_image)+'.jpg'
testimage = Image.open(path_image)
test_image = imread(path_image)
start_time = time.time()
for r in range(0,test_image.shape[0] - windowsize_r+1, 6):
def groundtruth():
global panelC, count
txt_path = path.split("/")[6]
txt_path_1 = txt_path.split(".")[0]
save_path = './' + str(txt_path_1) + '.txt'
txt_path = './2012image/' + str(txt_path_1) + '.txt'
read_path = './2012image/' + str(txt_path_1) + '.txt'
if not os.path.exists(read_path):
tk.messagebox.showerror(title='error message',
message='No such file or directory')
else:
count = 0
true_length = image.shape[1]
true_width = image.shape[0]
#read the orginal demension of the image
txtfile = open(txt_path)
firstline = txtfile.readlines()[0].split(",")
original_length = int(firstline[2])
original_width = int(firstline[3])
txtfile.close()
#read the number of the random points in the image
txtfile = open(txt_path)
count_points = int(txtfile.readlines()[5])
txtfile.close()
txtfile = open(txt_path)
data = txtfile.readlines()[6:6 + count_points]
corrdinate = np.zeros([count_points, 2], dtype=np.int)
for n in range(count_points):
data1 = data[n].split(",")
corrdinate[n,
0] = int(int(data1[0]) * true_length / original_length)
corrdinate[n, 1] = int(int(data1[1]) * true_width / original_width)
txtfile.close()
txtfile = open(txt_path)
label_encode = np.zeros(count_points)
label = txtfile.readlines()[6 + count_points:6 + count_points +
count_points]
for m in range(count_points):
label1 = label[m].split(",")
label2 = label1[1]
new_l = label2.replace('\"', '')
#coral
if ((new_l == "Agalg") or (new_l == "Aga") or (new_l == "Agaf")
or (new_l == "Col") or (new_l == "Helc")
or (new_l == "Mdrc") or (new_l == "Mdrcb")
or (new_l == "Mdrsd") or (new_l == "Mdrsf")
or (new_l == "Man") or (new_l == "Mon") or (new_l == "Ocu")
or (new_l == "Sco") or (new_l == "Sol") or (new_l == "Ste")
or (new_l == "STY")):
label_encode[m] = 0
#dead coral plate
elif (new_l == "DCP"):
label_encode[m] = 1
#rock
elif (new_l == "ROC"):
label_encode[m] = 2
#red alage
elif ((new_l == "CCA") or (new_l == "Amph") or (new_l == "Bot")
or (new_l == "Haly") or (new_l == "Kal") or (new_l == "Mar")
or (new_l == "PEY") or (new_l == "RHO")
or (new_l == "RHbl")):
label_encode[m] = 3
#green alage
elif ((new_l == "Ana") or (new_l == "Cau") or (new_l == "Caup")
or (new_l == "Caur") or (new_l == "Caus")
or (new_l == "Chae") or (new_l == "CHL") or (new_l == "Cod")
or (new_l == "Codin") or (new_l == "Hal")
or (new_l == "Halc") or (new_l == "Hald")
or (new_l == "Halt") or (new_l == "Micr")
or (new_l == "Ulva") or (new_l == "Venv")
or (new_l == "Verp")):
label_encode[m] = 4
else:
label_encode[m] = 5
txtfile.close()
crop_length = 30
crop_width = 30
all_image = np.zeros([count_points, crop_length, crop_width, 3],
dtype=np.uint8)
crop_x = int(crop_length / 2)
crop_y = int(crop_width / 2)
for i in range(count_points):
if (corrdinate[i, 0] - crop_x < 0):
corrdinate[i, 0] = crop_x
elif (corrdinate[i, 1] - crop_y < 0):
corrdinate[i, 1] = crop_y
elif (corrdinate[i, 0] + crop_x > true_length):
corrdinate[i, 0] = true_length - crop_x
elif (corrdinate[i, 1] + crop_y > true_width):
corrdinate[i, 1] = true_width - crop_y
all_image[i, :, :, :] = image[corrdinate[i, 1] -
crop_y:corrdinate[i, 1] + crop_y,
corrdinate[i, 0] -
crop_x:corrdinate[i, 0] + crop_x]
model = cnn_model()
model.summary()
model.load_weights('./model_weight/2012images-areas-7.5-50epoch.h5')
predict = model.predict(all_image, verbose=1)
res = np.argmax(predict, axis=1)
result_name = str(save_path) + ' points.jpg'
save_dir = os.path.join(os.getcwd(), '50point_result')
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
result_path = os.path.join(save_dir, result_name)
plt.imshow(image)
for i in range(50):
x = corrdinate[i, 0]
y = corrdinate[i, 1]
true_label = label_encode[i]
if (label_encode[i] == res[i]):
count = count + 1
mark_point(x, y, true_label)
else:
count = count
mark_point(x, y, true_label)
mark_wrong_point(x, y)
fig = plt.gcf()
fig.set_size_inches(image.shape[1] / 100.0 / 3.0,
image.shape[0] / 100.0 / 3.0)
plt.gca().xaxis.set_major_locator(plt.NullLocator())
plt.gca().yaxis.set_major_locator(plt.NullLocator())
plt.subplots_adjust(top=1,
bottom=0,
left=0,
right=1,
hspace=0,
wspace=0)
plt.margins(0, 0)
plt.savefig(result_path, bbox_inches='tight')
acc = ((count + 5) / 50) * 100
res_image = imread(result_path)
pil_res_image = Image.fromarray(res_image)
pil_res_image_resize = resize(800, 800, pil_res_image)
# ...and then to ImageTk format
res_image_show = ImageTk.PhotoImage(pil_res_image_resize)
if panelC is None: #or panelB is None:
# the first panel will store our original image
panelC = tk.Label(image=res_image_show)
panelC.image = res_image_show
panelC.place(x=300, y=100, anchor='nw')
# otherwise, update the image panels
else:
# update the pannels
panelC.configure(image=res_image_show)
panelC.image = res_image_show
panelC = None #very import to clean panel
var.set('Accuracy is: ' + str(acc) + '%' +
'\n X: mismatching \n the manual label')
test = np.uint8(testimage)
pixel = np.zeros([1, 1, 3], dtype=np.uint8)
for r in range(0, test.shape[0]):
for c in range(0, test.shape[1]):
pixel = test[r, c, :]
#if(np.all(pixel == blue)):
if ((pixel[0] == 0) and (pixel[1] == 0)):
count = count + 1
all_pixel = test.shape[0] * test.shape[1]
coral_percent = (count / all_pixel) * 100
coral_percent = round(coral_percent, 2)
return coral_percent
#%%load model and weight
model = cnn_model()
model.summary()
#model.load_weights('CNN_weights-08.h5')
model.load_weights('./saved_models/2012images_trained_augmentation-6.6.h5')
#model.load_weights('./saved_models/2012images_trained_model.h5')
#model.load_weights('./saved_models/cifar10_ResNet14v1_model.007.h5')
#model.load_weights('./saved_models/2012images_data_augmentation_trained_model-50epoch.h5')
#%%
windowsize_r = 30
windowsize_c = 30
raw_image_files = glob.glob("./2013-image/*.jpg")
count = 0
for name in raw_image_files:
name_piece = name.split("\\")[1]
name_image = name_piece.split(".")[0]
path_image = './2013-image/' + str(name_image) + '.jpg'
# load dataset
path_data = 'train/image'
path_truth = 'train/gtruth'
path_loss = 'train/boundary'
data_set = np.zeros((100, 512, 512))
target_set = np.zeros((100, 512, 512))
boundary_set = np.zeros((100, 512, 512))
for i in range(0, 100):
# read the information in grayscale images
data_set[i] = cv2.imread(path_data + '/' + str(i) + '.bmp', 0)
target_set[i] = cv2.imread(path_truth + '/' + str(i) + '.bmp', 0)
boundary_set[i] = cv2.imread(path_loss + '/' + str(i) + '.bmp', 0)
with tf.variable_scope('model_defination') as scope:
model_output = tf.cast(cnn_model(train_input, batch_size),
tf.float32,
name='train_output')
scope.reuse_variables()
test_output = tf.cast(cnn_model(test_input, 1),
tf.float32,
name='test_output')
loss = loss_function(model_output, train_target, boun_target)
accuracy = accuracy_of_batch(test_output, test_target)
generation_num = tf.Variable(0, trainable=False)
#geneartion_num get the current generation_number
#learning_rate initial 0.002
train_op = train_step(loss, generation_num, learning_rate,
