def gui(adr=None, scale=1, img_rows=720, img_cols=576):
"""
gui is a full gui program that takes an image as input, and gives the user the option to make its own mask
"""
channels = 3
"""
WARNING; here img shape is 720,576. this is trasposed of the rest of the program
"""
img_shape = (img_rows, img_cols, channels)
img, img_path = pl.load_one_img((img_cols, img_rows, channels), dest=adr)
scipy.misc.toimage(img, cmin=-1, cmax=1).save('tmp.png')
#starting window, and adding buttons
pygame.init()
button0 = pygame.Rect(img_rows + 25, 0, 50, 50)
button1 = pygame.Rect(img_rows + 25, 100, 50, 50)
button2 = pygame.Rect(img_rows + 25, 200, 50, 50)
button3 = pygame.Rect(img_rows + 25, 300, 50, 50)
buttons = [button0, button1, button2, button3]
global going
going = True
while going:
img, img_path = pl.load_one_img(img_shape, dest='tmp.png')
screen, px = _gui_setup('tmp.png', 1, buttons)
mask_array, going = _gui_mainLoop(screen,
px,
buttons,
img_shape=img_shape)
scipy.misc.toimage(mask_array.T, cmin=-1, cmax=1).save('tmp.png')
pygame.quit()
return mask_array
def main(adr='2.jpg', scale=1, img_cols=576, img_rows=720):
channels = 3
img_shape = (img_cols, img_rows, channels)
generator = load_model("saved_model/generator.h5")
generator.load_weights("saved_model/generator_weigths.h5")
img, img_path = pl.load_one_img(img_shape, '1.jpg')
scipy.misc.toimage(img, cmin=-1, cmax=1).save('working_img.png')
global going
going = True
for i in range(3):
mask = gui(adr='working_img.png').T
#import matplotlib.pyplot as plt
#plt.imshow(0.5*mask+0.5)
#plt.show()
np.save('/tmp/tmp_mask.npy', mask)
img, img_path = pl.load_one_img(img_shape, "working_img.png")
masked_imgs, missing_parts, m = mask_from_template(img,
'/tmp',
rot=False,
fliplr=False,
flipud=False)
gen_fake = generator.predict(missing_parts)
gen_fake = combine_imgs_with_mask(gen_fake, img, m)
gen_fake = np.squeeze(gen_fake, axis=0)
scipy.misc.toimage(gen_fake, cmin=-1, cmax=1).save('working_img.png')
def get_coord_live(adr='2.jpg',
scale=1,
img_rows=576,
img_cols=720,
mask_height=208,
mask_width=280):
"""
Only used for contextencoder/live.py
"""
channels = 3
img_shape = (img_rows, img_cols, channels)
missing_shape = (mask_width, mask_height, channels)
from keras.models import load_model
generator = load_model("saved_model/generator.h5")
generator.load_weights("saved_model/generator_weigths.h5")
import scipy.misc
import plotload as pl
import numpy as np
img, img_path = pl.load_one_img(img_shape, dest=adr)
scipy.misc.toimage(img, cmin=-1, cmax=1).save('tmp.png')
pygame.init()
global going
going = True
while going:
img, img_path = pl.load_one_img(img_shape, dest='tmp.png')
screen, px = _setup('tmp.png', 1)
left, upper, right, lower, going = _mainLoop(screen, px,
(mask_width, mask_height))
if not going:
scipy.misc.toimage(masked_img, cmin=-1, cmax=1).save('output.png')
break
(x1, x2, y1, y2) = (int(left * scale), int(upper * scale),
int(right * scale), int(lower * scale))
masked_img = np.empty_like(img)
missing_parts = np.ndarray(shape=(x2 - x1, y2 - y1))
masked_img = img.copy()
missing_parts = masked_img[y1:y2, x1:x2, :].copy()
masked_img[y1:y2, x1:x2, :] = -1
guess = np.squeeze(generator.predict(np.expand_dims(masked_img,
axis=0)),
axis=0)
masked_img[y1:y2, x1:x2, :] = guess
scipy.misc.toimage(masked_img, cmin=-1, cmax=1).save('tmp.png')
pygame.display.quit()
def train(self):
a = auto.CCgan(self.img_cols, self.img_rows)
#a.build_model()
#a.train_model()
a.load_model()
a.load_model_weights()
prepro = a.build_wrapper()
import plotload
test = plotload.load_one_img((256, 256), dest='green', crop=True)[0]
test2 = prepro(test)
train_datagen = ImageDataGenerator(rescale=1. / 255,
horizontal_flip=True,
vertical_flip=True,
fill_mode="nearest",
zoom_range=0.3,
width_shift_range=0.3,
height_shift_range=0.3,
rotation_range=30,
validation_split=0.1,
preprocessing_function=prepro)
val_datagen = ImageDataGenerator(rescale=1. / 255,
validation_split=0.9)
train_generator = train_datagen.flow_from_directory(
self.train_dir,
target_size=(self.img_cols, self.img_rows),
batch_size=self.batch_size,
class_mode="categorical")
val_generator = train_datagen.flow_from_directory(
self.train_dir,
target_size=(self.img_cols, self.img_rows),
batch_size=self.batch_size,
class_mode="categorical")
path, dirs, files = next(os.walk('./logs'))
file_count = len(files)
#checkpoint = ModelCheckpoint(f"vgg16_{file_count}.h5", monitor='val_acc', verbose=1, save_best_only=True, save_weights_only=True, mode='auto', period=1)
#checkpoint = ModelCheckpoint(f"densenet_{file_count}.h5", monitor='val_acc', verbose=1, save_best_only=True, save_weights_only=True, mode='auto', period=1)
checkpoint = ModelCheckpoint(f"resNet50_{file_count}.h5",
monitor='val_acc',
verbose=1,
save_best_only=True,
save_weights_only=True,
mode='auto',
period=1)
early = EarlyStopping(monitor='val_acc',
min_delta=0,
patience=5,
verbose=1,
mode='auto')
board = TensorBoard(f"./logs/run_{file_count}")
# Train the model
self.VGG.fit_generator(train_generator,
epochs=10,
validation_data=train_generator,
callbacks=[checkpoint, early, board])
def save_img(self,epoch):
test=plotload.load_one_img((256,256), dest='/home/mathias/Documents/kvasir-dataset-v2/med/stool-plenty/1.jpg',
extra_dim=True)
test2=A.model.predict(test)
plt.subplot(121)
plt.imshow(test[0]*0.5+0.5)
plt.subplot(122)
plt.imshow(test2[0]*0.5+0.5)
plt.savefig(f"epoc_{epoch}.jpg")
def fill(self, img_adr, img_shape):
import plotload as pl
img, img_path = pl.load_one_img(img_shape, dest=img_adr)
#WARNING; IMAGE IS NOT THE SAME AS SHOWN ON THE SCREEN, IT HAS BEEN CROPPED
masked_img, _, (y1, y2, x1, x2) = self.mask_select(img, img_path)
guess = np.squeeze(self.generator.predict(
np.expand_dims(masked_img, axis=0)),
axis=0)
masked_img[y1:y2, x1:x2, :] = guess
import scipy.misc
scipy.misc.toimage(masked_img, cmin=-1, cmax=1).save('outfile.png')
def sort_folder(self, w, path=None):
import os
import cv2
from tqdm import tqdm
from shutil import copyfile
import sys
if path is not None:
dirs_i = []
dirs_o = []
d = next(os.walk(path))[1]
for i in d:
if i == 'none' or i == 'green' or i == 'preprocessed':
continue
dirs_o.append(path + 'preprocessed/' + i)
dirs_i.append(path + i)
for i in dirs_o:
if not os.path.exists(i):
os.makedirs(i)
else:
polyps = 'polyps'
ulcerative_colitis = 'ulcerative-colitis'
dirs = [polyps, ulcerative_colitis]
if not os.path.exists(polyps_prep):
os.makedirs(polyps_prep)
if not os.path.exists(ulcerative_colitis_prep):
os.makedirs(ulcerative_colitis_prep)
for i, o in tqdm(zip(dirs_i, dirs_o)):
for img_name in os.listdir(i):
path = os.path.join(i, img_name)
#img=plotload.load_one_img((self.img_cols,self.img_rows), dest=path,
# extra_dim=True)
if False: #cutter.is_green(img):
tmp = cv2.imwrite(
os.path.join(o, img_name),
cv2.cvtColor(127.5 * w(img)[0] + 127.5,
cv2.COLOR_RGB2BGR))
else:
img = plotload.load_one_img(
(self.img_cols + 70, self.img_rows + 70),
dest=path,
extra_dim=True)
img = img[0, 10:self.img_cols + 70 - 60,
60:self.img_rows + 70 - 10]
tmp = cv2.imwrite(
os.path.join(o, img_name),
cv2.cvtColor(127.5 * img + 127.5, cv2.COLOR_RGB2BGR))
def save_img(self, epoch):
test = plotload.load_one_img(
(256, 256),
dest=
'/home/mathias/Documents/kvasir-dataset-v2/med/stool-plenty/1.jpg',
extra_dim=True)
test2, _, _ = ms.mask_green_corner(test.copy(), val=-1)
test2 = self.generator.predict(test2)
plt.subplot(121)
plt.imshow(test[0] * 0.5 + 0.5)
plt.subplot(122)
plt.imshow(test2[0] * 0.5 + 0.5)
plt.savefig(f"epoc_{epoch}.jpg")
plt.savefig(f"tmp.jpg")
def __init__(self,img_cols,img_rows):
"""
Initializes the autoencoder.
"""
self.set_training_info()
globals().update(self.info)
self.threshold=threshold
self.img_cols = img_cols#256 # Original is ~576
self.img_rows = img_rows#256 # Original is ~720
self.channels = 3 # RGB
self.img_shape=(self.img_cols,self.img_rows,self.channels)
dummy=plotload.load_one_img(self.img_shape, dest='med/green',extra_dim=True)
self.dims =cutter.find_square_coords(dummy)
self.model=None
self.pretrained=False