add all the python files

Author: kylekma

.gitignore

@@ -1,6 +1,7 @@
 # Byte-compiled / optimized / DLL files
 __pycache__/
-
+*
+!*.py
 *$py.class
 
 # C extensions
@@ -14,7 +15,6 @@ dist/
 downloads/
 eggs/
 .eggs/
-lib/
 lib64/
 parts/
 sdist/

3DGAN/lib/__init__.py

@@ -0,0 +1,308 @@
+# ------------------------------------------------------------------------------
+# Copyright (c) Tencent
+# Licensed under the GPLv3 License.
+# Created by Kai Ma ([email protected])
+# ------------------------------------------------------------------------------
+
+from __future__ import print_function
+from __future__ import absolute_import
+from __future__ import division
+
+from easydict import EasyDict
+import os
+import numpy as np
+
+__C = EasyDict()
+cfg = __C
+
+# Model Path
+__C.MODEL_SAVE_PATH = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..', 'save_models'))
+__C.CT_MIN_MAX = [0, 5800]
+__C.XRAY1_MIN_MAX = [0, 1200]
+__C.XRAY2_MIN_MAX = [0, 1700]
+__C.CT_MEAN_STD = [0., 1.0]
+__C.XRAY1_MEAN_STD = [0., 1.0]
+__C.XRAY2_MEAN_STD = [0., 1.0]
+
+'''
+Network
+  Generator
+'''
+__C.NETWORK = EasyDict()
+# of input image channels
+__C.NETWORK.input_nc_G = 3
+# of output image channels
+__C.NETWORK.output_nc_G = 3
+# of gen filters in first conv layer
+__C.NETWORK.ngf = 64
+# selects model to use for netG
+__C.NETWORK.which_model_netG = 'resnet_generator'
+# instance normalization or batch normalization
+__C.NETWORK.norm_G = 'instance'
+# no dropout for the generator
+__C.NETWORK.no_dropout = False
+# network initialization [normal|xavier|kaiming|orthogonal]
+__C.NETWORK.init_type = 'normal'
+# gan, lsgan, wgan, wgan_gp
+__C.NETWORK.ganloss = 'lsgan'
+# down sampling
+__C.NETWORK.n_downsampling = 3
+__C.NETWORK.n_blocks = 9
+# activation
+__C.NETWORK.activation_type = 'relu'
+
+'''
+Network
+  Discriminator
+'''
+# of input image channels
+__C.NETWORK.input_nc_D = 3
+# of output image channels
+# __C.NETWORK.output_nc_D = 1
+# of discrim filters in first conv layer
+__C.NETWORK.ndf = 64
+# selects model to use for netD
+__C.NETWORK.which_model_netD = 'basic'
+# only used if which_model_netD==n_layers, dtype = int
+__C.NETWORK.n_layers_D = 3
+# instance normalization or batch normalization, dtype = str
+__C.NETWORK.norm_D = 'instance3d'
+# output channels of discriminator network, dtype = int
+__C.NETWORK.n_out_ChannelsD = 1
+__C.NETWORK.pool_size = 50
+__C.NETWORK.if_pool = False
+__C.NETWORK.num_D = 3
+# add condition to discriminator network
+__C.NETWORK.conditional_D = False
+
+# of input image channels
+__C.NETWORK.map_input_nc_D = 1
+# of discrim filters in first conv layer
+__C.NETWORK.map_ndf = 64
+# selects model to use for netD
+__C.NETWORK.map_which_model_netD = 'basic'
+# only used if which_model_netD==n_layers
+__C.NETWORK.map_n_layers_D = 3
+# instance normalization or batch normalization, dtype = str
+__C.NETWORK.map_norm_D = 'instance'
+# output channels of discriminator network, dtype = int
+__C.NETWORK.map_n_out_ChannelsD = 1
+__C.NETWORK.map_pool_size = 50
+__C.NETWORK.map_num_D = 3
+
+'''
+Train
+'''
+__C.TRAIN = EasyDict()
+# initial learning rate for adam
+__C.TRAIN.lr = 0.0002
+# momentum term of adam
+__C.TRAIN.beta1 = 0.5
+__C.TRAIN.beta2 = 0.9
+# if true, takes images in order to make batches, otherwise takes them randomly
+__C.TRAIN.serial_batches = False
+__C.TRAIN.batch_size = 1
+# threads for loading data
+__C.TRAIN.nThreads = 5
+# __C.TRAIN.max_epoch = 10
+# learning rate policy: lambda|step|plateau
+__C.TRAIN.lr_policy = 'lambda'
+# of iter at starting learning rate
+__C.TRAIN.niter = 100
+# of iter to linearly decay learning rate to zero
+__C.TRAIN.niter_decay = 100
+# multiply by a gamma every lr_decay_iters iterations
+__C.TRAIN.lr_decay_iters = 50
+# frequency of showing training results on console
+__C.TRAIN.print_freq = 10
+# frequency of showing training results on console
+__C.TRAIN.print_img_freq = 200
+# save model
+__C.TRAIN.save_latest_freq = 3000
+# save model frequent
+__C.TRAIN.save_epoch_freq = 5
+__C.TRAIN.begin_save_epoch = 0
+
+__C.TRAIN.weight_decay_if = False
+
+'''
+TEST
+'''
+__C.TEST = EasyDict()
+__C.TEST.howmany_in_train = 10
+
+'''
+Data
+Augmentation
+'''
+__C.DATA_AUG = EasyDict()
+__C.DATA_AUG.select_slice_num = 0
+__C.DATA_AUG.fine_size = 256
+__C.DATA_AUG.ct_channel = 256
+__C.DATA_AUG.xray_channel = 1
+__C.DATA_AUG.resize_size = 289
+
+'''
+2D GAN define loss
+'''
+__C.TD_GAN = EasyDict()
+# identity loss
+__C.TD_GAN.idt_lambda = 10.
+__C.TD_GAN.idt_reduction = 'elementwise_mean'
+__C.TD_GAN.idt_weight = 0.
+__C.TD_GAN.idt_weight_range = [0., 1.]
+__C.TD_GAN.restruction_loss = 'l1'
+# perceptual loss
+__C.TD_GAN.fea_m_lambda = 10.
+# output of discriminator
+__C.TD_GAN.discriminator_feature = True
+# wgan-gp
+__C.TD_GAN.wgan_gp_lambda = 10.
+# identity loss of map
+__C.TD_GAN.map_m_lambda = 0.
+# 'l1' or 'mse'
+__C.TD_GAN.map_m_type = 'l1'
+__C.TD_GAN.fea_m_map_lambda = 10.
+# Discriminator train times
+__C.TD_GAN.critic_times = 1
+
+'''
+3D GD-GAN define structure
+'''
+__C.D3_GAN = EasyDict()
+__C.D3_GAN.noise_len = 1000
+__C.D3_GAN.input_shape = [4,4,4]
+# __C.D3_GAN.input_shape_nc = 512
+__C.D3_GAN.output_shape = [128,128,128]
+# __C.D3_GAN.output_shape_nc = 1
+__C.D3_GAN.encoder_input_shape = [128, 128]
+__C.D3_GAN.encoder_input_nc = 1
+__C.D3_GAN.encoder_norm = 'instance'
+__C.D3_GAN.encoder_blocks = 4
+__C.D3_GAN.multi_view = [1,2,3]
+__C.D3_GAN.min_max_norm = False
+__C.D3_GAN.skip_number = 1
+# DoubleBlockLinearUnit Activation [low high k]
+__C.D3_GAN.dblu = [0., 1.0, 1.0]
+
+'''
+CT GAN
+'''
+__C.CTGAN = EasyDict()
+# input x-ray direction, 'H'-FrontBehind 'D'-UpDown 'W'-LeftRight
+# 'HDW' Means that deepness is 'H' and projection in plane of 'DW'
+#  relative to CT.
+__C.CTGAN.Xray1_Direction = 'HDW'
+__C.CTGAN.Xray2_Direction = 'WDH'
+# dimension order of input CT is 'DHW'(should add 'NC'-01 to front when training)
+__C.CTGAN.CTOrder = [0, 1, 2, 3, 4]
+# NCHDW to xray1 and NCWDH to xray2
+__C.CTGAN.CTOrder_Xray1 = [0, 1, 3, 2, 4]
+__C.CTGAN.CTOrder_Xray2 = [0, 1, 4, 2, 3]
+# identity loss'weight
+__C.CTGAN.idt_lambda = 1.0
+__C.CTGAN.idt_reduction = 'elementwise_mean'
+__C.CTGAN.idt_weight = 0.
+__C.CTGAN.idt_weight_range = [0., 1.]
+# 'l1' or 'mse'
+__C.CTGAN.idt_loss = 'l1'
+# feature metrics loss
+__C.CTGAN.feature_D_lambda = 0.
+# projection loss'weight
+__C.CTGAN.map_projection_lambda = 0.
+# 'l1' or 'mse'
+__C.CTGAN.map_projection_loss = 'l1'
+# gan loss'weight
+__C.CTGAN.gan_lambda = 1.0
+# multiView GAN auxiliary loss
+__C.CTGAN.auxiliary_lambda = 0.
+# 'l1' or 'mse'
+__C.CTGAN.auxiliary_loss = 'mse'
+# map discriminator
+__C.CTGAN.feature_D_map_lambda = 0.
+__C.CTGAN.map_gan_lambda = 1.0
+
+def cfg_from_yaml(filename):
+  '''
+  Load a config file and merge it into the default options
+  :param filename:
+  :return:
+  '''
+  import yaml
+  with open(filename, 'r') as f:
+    yaml_cfg = EasyDict(yaml.load(f))
+  _merge_a_into_b(yaml_cfg, __C)
+
+def print_easy_dict(easy_dict):
+  print('==='*10)
+  print('====YAML Parameters')
+  for k,v in easy_dict.__dict__.items():
+    print('{}: {}'.format(k, v))
+  print('==='*10)
+
+def merge_dict_and_yaml(in_dict, easy_dict):
+  if type(easy_dict) is not EasyDict:
+    return in_dict
+  easy_list = _easy_dict_squeeze(easy_dict)
+  for (k, v) in easy_list:
+    if k in in_dict:
+      raise KeyError('The same Key appear {}/{}'.format(k,k))
+  out_dict = EasyDict(dict(easy_list + list(in_dict.items())))
+  return out_dict
+
+def _easy_dict_squeeze(easy_dict):
+  if type(easy_dict) is not EasyDict:
+    print('Not EasyDict!!!')
+    return []
+
+  total_list = []
+  for k, v in easy_dict.items():
+    # recursively merge dicts
+    if type(v) is EasyDict:
+      try:
+        total_list += _easy_dict_squeeze(v)
+      except:
+        print('Error under config key: {}'.format(k))
+        raise
+    else:
+      total_list.append((k, v))
+  return total_list
+
+def _merge_a_into_b(a, b):
+  '''
+  Merge easyDict a to easyDict b
+  :param a: from easyDict
+  :param b: to easyDict
+  :return:
+  '''
+  if type(a) is not EasyDict:
+    return
+
+  for k, v in a.items():
+    # check k in a or not
+    if k not in b:
+      raise KeyError('{} is not a valid config key'.format(k))
+
+    old_type = type(b[k])
+    if old_type is not type(v):
+      if isinstance(b[k], np.ndarray):
+        v = np.array(v, dtype=b[k].type)
+      else:
+        raise ValueError('Type mismatch ({} vs. {})'
+                         'for config key: {}'.format(type(b[k]), type(v), k))
+    # recursively merge dicts
+    if type(v) is EasyDict:
+      try:
+        _merge_a_into_b(a[k], b[k])
+      except:
+        print('Error under config key: {}'.format(k))
+        raise
+    else:
+      b[k] = v
+
+
+
+
+
+
+

3DGAN/lib/config/__init__.py

@@ -0,0 +1,73 @@
+# ------------------------------------------------------------------------------
+# Copyright (c) Tencent
+# Licensed under the GPLv3 License.
+# Created by Kai Ma ([email protected])
+# ------------------------------------------------------------------------------
+
+from __future__ import print_function
+from __future__ import absolute_import
+from __future__ import division
+
+from lib.dataset.baseDataSet import Base_DataSet
+from lib.dataset.utils import *
+import h5py
+import numpy as np
+
+
+class AlignDataSet(Base_DataSet):
+  '''
+  DataSet For unaligned data
+  '''
+  def __init__(self, opt):
+    super(AlignDataSet, self).__init__()
+    self.opt = opt
+    self.ext = '.h5'
+    self.dataset_paths = get_dataset_from_txt_file(self.opt.datasetfile)
+    self.dataset_paths = sorted(self.dataset_paths)
+    self.dataset_size = len(self.dataset_paths)
+    self.dir_root = self.get_data_path
+    self.data_augmentation = self.opt.data_augmentation(opt)
+
+  @property
+  def name(self):
+    return 'AlignDataSet'
+
+  @property
+  def get_data_path(self):
+    path = os.path.join(self.opt.dataroot)
+    return path
+
+  @property
+  def num_samples(self):
+    return self.dataset_size
+
+  def get_image_path(self, root, index_name):
+    img_path = os.path.join(root, index_name, 'ct_xray_data'+self.ext)
+    assert os.path.exists(img_path), 'Path do not exist: {}'.format(img_path)
+    return img_path
+
+  def load_file(self, file_path):
+    hdf5 = h5py.File(file_path, 'r')
+    ct_data = np.asarray(hdf5['ct'])
+    x_ray1 = np.asarray(hdf5['xray1'])
+    x_ray1 = np.expand_dims(x_ray1, 0)
+    hdf5.close()
+    return ct_data, x_ray1
+
+  '''
+  generate batch
+  '''
+  def pull_item(self, item):
+    file_path = self.get_image_path(self.dir_root, self.dataset_paths[item])
+    ct_data, x_ray1 = self.load_file(file_path)
+
+    # Data Augmentation
+    ct, xray1 = self.data_augmentation([ct_data, x_ray1])
+
+    return ct, xray1, file_path
+
+
+
+
+
+