def flow_from_directory(self, directory, mask_generator=MaskGenerator(256,256), *args, **kwargs):
generator = super().flow_from_directory(directory, class_mode=None, *args, **kwargs)
seed = None if 'seed' not in kwargs else kwargs['seed']
while True:
original_image = next(generator)
mask = np.stack([mask_generator.sample(seed) for _ in range(original_image.shape[0])], axis=0)
masked_image = deepcopy(original_image)
masked_image *= mask
yield [masked_image, mask], original_image
import cv2
import numpy as np
import matplotlib.pyplot as plt
from copy import deepcopy
from libs.util import MaskGenerator, torch_preprocessing, torch_postprocessing
from libs.DataGenerator import MaskedDataGenerator
# mask generation example
generator = MaskGenerator(256, 256, 3)
# get mask
mask = generator.sample()
# display with opencv
cv2.imshow('mask displayed with opencv', mask * 255)
print("Cheeck")
# get many masks
masks = np.stack([generator.sample() for _ in range(3)], axis=0)
# display them with matplotlib
fig, axes = plt.subplots(1, 3, figsize=(10, 4))
axes[0].imshow(masks[0, :, :, :] * 255)
axes[1].imshow(masks[1, :, :, :] * 255)
axes[2].imshow(masks[2, :, :, :] * 255)
axes[0].set_title('mask 0')
axes[1].set_title('mask 1')
axes[2].set_title('mask 2')
fig.suptitle('Masks displayed with matplotlib', fontsize=12)
# print(masked.shape, ori.shape)
gc.collect()
yield [masked, mask], ori
# Create training generator
train_datagen = AugmentingDataGenerator(
rotation_range=10,
width_shift_range=0.1,
height_shift_range=0.1,
rescale=1./255,
horizontal_flip=True
)
train_generator = train_datagen.flow_from_directory(
TRAIN_DIR,
MaskGenerator(400, 400, 3),
target_size=(400, 400),
batch_size=BATCH_SIZE
)
# Create validation generator
val_datagen = AugmentingDataGenerator(rescale=1./255)
val_generator = val_datagen.flow_from_directory(
VAL_DIR,
MaskGenerator(400, 400, 3),
target_size=(400, 400),
batch_size=BATCH_SIZE,
#classes=['val'],
)
# Create testing generator
args = parse_args()
if args.stage == 'finetune' and not args.checkpoint:
raise AttributeError('If you are finetuning your model, you must supply a checkpoint file')
# Create training generator
train_datagen = AugmentingDataGenerator(
rotation_range=10,
width_shift_range=0.1,
height_shift_range=0.1,
rescale=1./255,
horizontal_flip=True
)
train_generator = train_datagen.flow_from_directory(
args.train,
MaskGenerator(512, 512, 3),
target_size=(512, 512),
batch_size=args.batch_size
)
# Create validation generator
val_datagen = AugmentingDataGenerator(rescale=1./255)
val_generator = val_datagen.flow_from_directory(
args.validation,
MaskGenerator(512, 512, 3),
target_size=(512, 512),
batch_size=args.batch_size,
classes=['val'],
seed=42
)
import numpy as np import cv2 from libs.util import MaskGenerator, ImageChunker mask = MaskGenerator(128, 128, 3, rand_seed=1222)._generate_mask() mask = mask[0:63, 0:63, 0] # import keras.activations as activations # import tensorflow as tf # f = np.array([[1, 2, 1], # [1, 0, 0], # [-1, 0, 1]]) # img = np.array([ # [2, 3, 7, 4, 6, 2, 9], # [6, 6, 9, 8, 7, 4, 3], # [3, 4, 8, 3, 8, 9, 7], # [7, 8, 3, 6, 6, 3, 4], # [4, 2, 1, 8, 3, 4, 6], # [3, 2, 4, 1, 9, 8, 3], # [4, 5, 3, 9, 2, 1, 4]]) # img=np.random.randint(0,50,(63, 63)) 这是生成随机数字的矩阵 input_img = cv2.imread(r"C:\Users\dell\Desktop\paper2\\figure\Fig3\\img.jpg") # img0=np.array(input_img) img = input_img[:, :, 0] masked_img = img * mask kernel1 = np.random.rand(7, 7) kernel2 = np.random.rand(5, 5) kernel3 = np.random.rand(3, 3) mask_kernel = np.ones([7, 7]) # f=round(f,1) strde = 1
args = parse_args()
if args.stage == 'finetune' and not args.checkpoint:
raise AttributeError(
'If you are finetuning your model, you must supply a checkpoint file'
)
# Create training generator
train_datagen = AugmentingDataGenerator(rotation_range=10,
width_shift_range=0.1,
height_shift_range=0.1,
rescale=1. / 255,
horizontal_flip=True)
train_generator = train_datagen.flow_from_directory(
args.train,
MaskGenerator(512, 512, 3, filepath=args.masks),
target_size=(512, 512),
batch_size=args.batch_size)
# Create validation generator
val_datagen = AugmentingDataGenerator(rescale=1. / 255)
val_generator = val_datagen.flow_from_directory(args.validation,
MaskGenerator(
512,
512,
3,
filepath=args.masks),
target_size=(512, 512),
batch_size=args.batch_size,
classes=['val']
# seed=42
#model4.load(r"D:\PycharmProjects2\PConv-Keras\data\logs\Thanka_phase1\p1t55\weights.25-1.28.h5", train_bn=False)
#model5.load(r"D:\PycharmProjects2\PConv-Keras\data\logs\Thanka_phase1\p1t44\weights.40-0.18.h5", train_bn=False)
#model6.load(r"D:\PycharmProjects2\PConv-Keras\data\logs\Thanka_phase1\p1t36\weights.59-0.30.h5", train_bn=False)
# model1.load(r"D:\PycharmProjects2\PConv-Keras\data\logs\Thanka_phase1\p1t16\weights.11-1.15.h5", train_bn=False)
models = []
models.append(model1)
# models.append(model2)
#models.append(model3)
#models.append(model4)
#models.append(model5)
#models.append(model6)
mse = [0, 0, 0, 0]
psnr = [0, 0, 0, 0]
ssim = [0, 0, 0, 0]
image_num = 0
mask = MaskGenerator(512, 512, 3, rand_seed=4210)._generate_mask()
start_time = datetime.now()
for filename in os.listdir(original_img_folder):
image_num = image_num + 1
image = cv2.imread(os.path.join(original_img_folder, filename))
input_img = image
image = Image.fromarray(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
image = np.array(image) / 255
# mask = MaskGenerator(512, 512)._generate_mask()
image[mask == 0] = 1
for j in range(0, len(models)):
predicted_img = models[j].predict(
[np.expand_dims(image, 0),
np.expand_dims(mask, 0)])[0] * 255
# if you want to save inpainted result, please use the following 3 lines, else use the 4th line below:
for crop, ax in zip(crops, axes.flatten()):
# Load image
im = Image.open(SAMPLE_IMAGE).resize((2048, 2048))
# Crop image
h, w = im.height, im.width
left = np.random.randint(0, w - crop[1])
right = left + crop[1]
upper = np.random.randint(0, h - crop[0])
lower = upper + crop[0]
im = im.crop((left, upper, right, lower))
# Create masked array
im = np.array(im) / 255
mask_gen = MaskGenerator(*crop)
mask = mask_gen._generate_mask()
im[mask == 0] = 1
# Store for prediction
imgs.append(im)
masks.append(mask)
# Show image
ax.imshow(im)
ax.set_title("{}x{}".format(crop[0], crop[1]))
from libs.pconv_model import PConvUnet
model = PConvUnet(vgg_weights=None, inference_only=True)
model.load(r"/content/pconv_imagenet.26-1.07.h5", train_bn=False)
chunker = ImageChunker(512, 512, 30)
import os
from copy import deepcopy
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
import cv2
from libs.util import MaskGenerator
# for i in range (0,100):
# mask = MaskGenerator(512,512)._generate_mask()
# #mask = MaskGenerator(256,256)._generate_mask()
# # #mask = MaskGenerator(512, 512, 3, rand_seed = 666)._generate_mask()
# cv2.imwrite('./temp_mask/mask_temp_'+str(i)+'.png', mask*255)
mask = MaskGenerator(512, 512, 3, rand_seed=22445)._generate_mask()
cv2.imwrite('./temp_mask/mask_temp_20200407.png', mask * 255)
import cv2
import numpy as np
import matplotlib.pyplot as plt
from copy import deepcopy
from libs.util import MaskGenerator
# mask reduction in partial convolutions: example
generator = MaskGenerator(256, 256, 3)
# get mask
mask = generator.sample()
clusterSize = (2, 2)
fig, axes = plt.subplots(1, 4, figsize=(10, 5))
for step in range(3):
axes[step].imshow(mask * 255)
newmask = deepcopy(mask)
for i in range(clusterSize[0], 256 - clusterSize[0]):
for j in range(clusterSize[1], 256 - clusterSize[1]):
for k in range(i - clusterSize[0], i + clusterSize[0]):
for m in range(j - clusterSize[1], j + clusterSize[1]):
if mask[k, m][0] == 1:
newmask[i, j] = (1, 1, 1)
mask = newmask
axes[3].imshow(mask * 255)
plt.show()