Am学习使用深度学习（特别是U-Net模型）进行图像分割
该错误可能很小，但我无法触发
我已经尝试过多次运行此代码以进行前列腺图像分割。
我最近读过其他人发布的相同错误，但与我的错误不符。
我不知道该怎么解决。
Epoch 1/1
Exception in thread Thread-6:
Traceback (most recent call last):
File "C:\Users\exaud\Anaconda3\lib\threading.py", line 917, in _bootstrap_inner
self.run()
File "C:\Users\exaud\Anaconda3\lib\threading.py", line 865, in run
self._target(*self._args, **self._kwargs)
File "C:\Users\exaud\Anaconda3\lib\site-packages\keras\utils\data_utils.py", line 666, in _run
with closing(self.executor_fn(_SHARED_SEQUENCES)) as executor:
File "C:\Users\exaud\Anaconda3\lib\site-packages\keras\utils\data_utils.py", line 661, in <lambda>
initargs=(seqs, self.random_seed))
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\context.py", line 119, in Pool
context=self.get_context())
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\pool.py", line 177, in __init__
self._repopulate_pool()
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\pool.py", line 238, in _repopulate_pool
self._wrap_exception)
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\pool.py", line 257, in _repopulate_pool_static
w.start()
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\process.py", line 112, in start
self._popen = self._Popen(self)
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\context.py", line 322, in _Popen
return Popen(process_obj)
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\popen_spawn_win32.py", line 65, in __init__
reduction.dump(process_obj, to_child)
File "C:\Users\exaud\Anaconda3\lib\multiprocessing\reduction.py", line 60, in dump
ForkingPickler(file, protocol).dump(obj)
TypeError: can't pickle _thread.lock objects
该模型基于U-Net构建，并使用Keras和TensorFlow
这是用于训练该模型的代码
from __future__ import division, print_function
import numpy as np
print()
import pydicom
from collections import defaultdict
import os, sys, pickle
import shutil
import matplotlib.pyplot as plt
import nrrd
from skimage.transform import resize
from skimage.exposure import equalize_adapthist, equalize_hist
from keras.optimizers import Adam, SGD
from keras.callbacks import ModelCheckpoint
from keras.callbacks import LearningRateScheduler
from keras.preprocessing.image import ImageDataGenerator
from models import actual_unet, simple_unet
from metrics import dice_coef, dice_coef_loss, numpy_dice
from augmenters import elastic_transform
def dicom_to_array(img_rows, img_cols):
for direc in ['train', 'test', 'validate']:
fname = 'E:/Research/Codes/prostate_segmentation_u-net/data/' + direc + '_dict.pkl'
PathDicom = 'E:/Research/Codes/prostate_segmentation_u-net/data/' + direc + '/'
dcm_dict = dict() # create an empty list
for dirName, subdirList, fileList in os.walk(PathDicom):
if any(".dcm" in s for s in fileList):
ptn_name = dirName.split('/')[3]
fileList = filter(lambda x: '.dcm' in x, fileList)
indice = [ int( fname[:-4] ) for fname in fileList]
imgs = np.zeros( [indice[-1]+1, img_rows, img_cols])
for filename in np.sort(fileList):
img = pydicom.read_file(os.path.join(dirName,filename)).pixel_array.T
img = equalize_hist( img.astype(int) )
img = resize( img, (img_rows, img_cols), preserve_range=True)
imgs[int(filename[:-4])] = img
dcm_dict[ptn_name] = imgs
imgs = []
img_masks = []
for patient in dcm_dict.keys():
for fnrrd in os.listdir(PathDicom):
if fnrrd.startswith(patient) and fnrrd.endswith('nrrd'):
masks = np.rollaxis(nrrd.read(PathDicom + fnrrd)[0], 2)
rescaled = np.zeros( [ len(masks), img_rows, img_cols])
for mm in range(len(rescaled)):
rescaled[mm] = resize( masks[mm], (img_rows, img_cols), preserve_range=True)/2.0
masks = rescaled.copy()
#Check if the dimension of the masks and the images match
if len(dcm_dict[patient]) != len(masks) :
print('Dimension mismatch for', patient, 'in folder', direc)
else:
img_masks.append(masks)
imgs.append( dcm_dict[patient] )
break
imgs = np.concatenate((imgs,imgs),axis=0, out=None).reshape(-1, img_rows, img_cols, 1)
img_masks = np.concatenate((img_masks,img_masks),axis=0, out=None).reshape(-1, img_rows, img_cols, 1)
#I will do only binary classification for now
img_masks = np.array(img_masks>0.45, dtype=int)
np.save('E:/Research/Codes/prostate_segmentation_u-net/data/' + direc + '.npy', imgs)
np.save('E:/Research/Codes/prostate_segmentation_u-net/data/' + direc + '_masks.npy', img_masks)
def load_data():
X_train = np.load('E:/Research/Codes/prostate_segmentation_u-net/data/train.npy')
y_train = np.load('E:/Research/Codes/prostate_segmentation_u-net/data/train_masks.npy')
X_test = np.load('E:/Research/Codes/prostate_segmentation_u-net/data/test.npy')
y_test = np.load('E:/Research/Codes/prostate_segmentation_u-net/data/test_masks.npy')
X_val = np.load('E:/Research/Codes/prostate_segmentation_u-net/data/validate.npy')
y_val = np.load('E:/Research/Codes/prostate_segmentation_u-net/data/validate_masks.npy')
return X_train, y_train, X_test, y_test, X_val, y_val
# learning rate schedule
def step_decay(epoch):
initial_lrate = 0.001
drop = 0.5
epochs_drop = 5
lrate = initial_lrate * drop**int((1 + epoch) / epochs_drop)
return lrate
def keras_fit_generator(img_rows=96, img_cols=96, n_imgs=10**4, batch_size=32, regenerate=True):
if regenerate:
dicom_to_array(img_rows, img_cols)
#preprocess_data()
X_train, y_train, X_test, y_test, X_val, y_val = load_data()
img_rows = X_train.shape[1]
img_cols = img_rows
# we create two instances with the same arguments
data_gen_args = dict(
featurewise_center=False,
featurewise_std_normalization=False,
rotation_range=90.,
width_shift_range=0.1,
height_shift_range=0.1,
horizontal_flip=True,
vertical_flip=True,
zoom_range=0.2)#,
#preprocessing_function=elastic_transform)
image_datagen = ImageDataGenerator(**data_gen_args)
mask_datagen = ImageDataGenerator(**data_gen_args)
# Provide the same seed and keyword arguments to the fit and flow methods
seed = 1
image_datagen.fit(X_train,seed=seed)
mask_datagen.fit(y_train, seed=seed)
image_generator = image_datagen.flow(X_train, batch_size=batch_size, seed=seed)
mask_generator = mask_datagen.flow(y_train, batch_size=batch_size, seed=seed)
#from itertools import izip
#from itertools import zip_longest
try:
from itertools import izip
except ImportError: #python3.x
izip = zip
train_generator = zip(image_generator, mask_generator)
model = simple_unet( img_rows, img_cols)
model.load_weights('E:/Research/Codes/prostate_segmentation_u-net/data/weights.h5')
model.summary()
model_checkpoint = ModelCheckpoint(
'E:/Research/Codes/prostate_segmentation_u-net/data/weights.h5', monitor='val_loss', save_best_only=True)
lrate = LearningRateScheduler(step_decay)
model.compile( optimizer=Adam(), loss=dice_coef_loss, metrics=[dice_coef, 'binary_accuracy'])
model.fit_generator(
train_generator,
steps_per_epoch=n_imgs//batch_size,
epochs=1,
verbose=1,
shuffle=True,
validation_data=(X_val, y_val),
callbacks=[model_checkpoint,lrate],
use_multiprocessing=True
)
score = model.evaluate(X_test, y_test, verbose=2)
print()
print('Test accuracy:', score[1])
import time
start = time.time()
keras_fit_generator(img_rows=96, img_cols=96, regenerate=True,
n_imgs=15*10**4, batch_size=32)
end = time.time()
print('Elapsed time:', round((end-start)/60, 2 ) )