-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathmain_mcVAE.py
executable file
·170 lines (138 loc) · 7.68 KB
/
main_mcVAE.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
import numpy as np
import os
import torch
import pandas as pd
import argparse
import utils.io.image as io_func
from utils.sitk_np import np_to_sitk
from torchvision.utils import save_image
import pdb
import datetime
from shutil import copyfile
from dataloader.MM_loader import MM_loader
from mcvae import pytorch_modules, utilities, preprocessing, plot, diagnostics
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Fundus/CMR Modalities')
parser.add_argument('--n_channels', default=2, type=int) # number of channels for MCVAE
parser.add_argument('--lat_dim', default=2048, type=int)
parser.add_argument('--lr', type=float, default=5e-4)
parser.add_argument('--epochs', default=2000, type=int)
parser.add_argument('--save_model', default=200, type=int) # save the model every x epochs
parser.add_argument('--batch_size', default=8, type=int)
parser.add_argument('--n_cpu', default=24, type=int)
parser.add_argument('--dir_dataset', type=str, default='./input_data/')
parser.add_argument('--dir_ids', type=str, default='./input_data/ids/manual_LVM_LVEDV_mtdt.csv')
parser.add_argument('--sax_img_size', type=list, default=[128, 128, 15])
parser.add_argument('--fundus_img_size', type=int, default=128)
parser.add_argument('--ndf', type=int, default=128)
parser.add_argument('--test_mode', type=bool, default=True)
parser.add_argument('--save_test_imgs', type=bool, default=True)
parser.add_argument('--dir_test_ids', type=str, default='2020-05-17_11-36-05/')
parser.add_argument('--dir_results', type=str, default='./results/')
parser.add_argument('--percentage', type=float, default=0.90)
args = parser.parse_args()
# Multi-channel VAE config
init_dict = {
'n_channels': args.n_channels,
'lat_dim': args.lat_dim, # We fit args.lat_dim latent dimensions
'n_feats': {'fundus': [3, args.ndf, args.fundus_img_size],
'cmr': [args.sax_img_size[2], args.ndf, args.sax_img_size[0]]
},
'opt': args
}
if not args.test_mode:
args.dir_results = args.dir_results + datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S') + '/'
if not os.path.exists(args.dir_results):
os.makedirs(args.dir_results)
print('\nLoading IDs file\n')
IDs = pd.read_csv(args.dir_ids, sep=',')
# Dividing the number of images for training and test.
IDs_copy = IDs.copy()
train_set = IDs_copy.sample(frac = args.percentage, random_state=0)
test_set = IDs_copy.drop(train_set.index)
test_set.to_csv(args.dir_results + 'test_set.csv', index=False)
train_set.to_csv(args.dir_results + 'train_set.csv', index=False)
copyfile('main_mcVAE.py', args.dir_results + 'main_mcVAE.py')
copyfile('./networks/VAE_net.py', args.dir_results + 'VAE_net.py')
copyfile('./dataloader/MM_loader.py', args.dir_results + 'MM_loader.py')
train_loader = MM_loader(batch_size = args.batch_size,
fundus_img_size = args.fundus_img_size,
num_workers = args.n_cpu,
sax_img_size = args.sax_img_size,
shuffle = True,
dir_imgs = args.dir_dataset,
args = args,
ids_set = train_set
)
# Creating model
model_MM = pytorch_modules.MultiChannelSparseVAE(**init_dict)
model_MM.init_loss()
model_MM.optimizer = torch.optim.Adam(model_MM.parameters(), lr=args.lr, betas=(0.5, 0.999), weight_decay=1e-5)
# Optimizing model
model_MM.optimize(epochs = args.epochs, data = train_loader)
# Saving model
utilities.save_model(model_MM, filename= args.dir_results + 'model_' + str(args.epochs) + '_epochs_dict')
# Creating plots
print('Significant dimensions: ', model_MM.dropout.cpu().detach().numpy())
significant_dim = np.where(model_MM.dropout.cpu().detach().numpy()<0.5)[1]
# Plotting model convergence
diagnostics.plot_loss(model_MM, save_fig=True, path_plot = args.dir_results)
else:
print('\nTesting Mode. Loading IDs files \n')
# Reading the files that contains labels and names.
# test_set = pd.read_csv(args.dir_results + args.dir_test_ids + 'train_set.csv', sep=',')
test_set = pd.read_csv(args.dir_ids, sep=',') # This is used to reconstructed both train and test set
test_loader = MM_loader(batch_size = args.batch_size,
fundus_img_size = args.fundus_img_size,
num_workers = args.n_cpu,
sax_img_size = args.sax_img_size,
shuffle = False,
dir_imgs = args.dir_dataset,
args = args,
ids_set = test_set
)
# Loading model
print('Loading model ...')
# Creating model
model_MM = pytorch_modules.MultiChannelSparseVAE(**init_dict)
loaded_model = utilities.load_model(args.dir_results + args.dir_test_ids)
model_MM.load_state_dict(loaded_model['state_dict'])
print('Making predictions ...')
laten_vars_fundus = []
laten_vars_cmr = []
img_names_4_linear_reg = []
labels_4_linear_reg = []
for i, (fundus, sax, mtdt, img_names) in enumerate(test_loader):
fundus = fundus.cuda()
sax = sax.cuda()
print('Batch: ' + str(i))
# Getting predictions
inputToLatent = model_MM.encode((fundus, sax))
latent_vars = model_MM.sample_from(inputToLatent)
predictions = model_MM.decode(latent_vars)
############# For linear regression ############
# Running for the batch size
for l in range(len(img_names)):
# concatenating latent variables and demographic
laten_vars_fundus.append(np.concatenate([latent_vars[0].cpu().detach().numpy()[l], mtdt.cpu().detach().numpy()[l][2:]]))
laten_vars_cmr.append(np.concatenate([latent_vars[1].cpu().detach().numpy()[l], mtdt.cpu().detach().numpy()[l][2:]]))
labels_4_linear_reg.append(mtdt.cpu().detach().numpy()[l][:2])
img_names_4_linear_reg.append(int(img_names[l].split('_')[0]))
###################################################
# Dir for the generated data
gen_dir = args.dir_results + args.dir_test_ids + 'gen_data/'
if not os.path.exists(gen_dir):
os.makedirs(gen_dir)
for d in range(len(img_names)):
# Saving image result
if args.save_test_imgs:
n = min(fundus.size(0), 8)
comparison = torch.cat([fundus[:n], predictions[0][0].loc[:n]])
# save_image(comparison.cpu(), gen_dir + 'fundus_generated_{}'.format(img_names[:n]) + '.png', nrow=n)
# save_image(predictions[0][0].loc[d], gen_dir + 'fundus_generated_' + img_names[d] + '.png')
for idx in range(fundus.size(0)):
name_cmr = img_names[idx].split('_')[0]
io_func.write(np_to_sitk(predictions[0][1].loc.cpu().detach().numpy()[idx]), gen_dir + 'reconstructed_' + name_cmr + '.vtk')
from utils.linear_reg import linear_reg
linear_reg(laten_vars_fundus, labels_4_linear_reg, img_names_4_linear_reg, 'fundus', args)
# linear_reg(laten_vars_cmr, labels_4_linear_reg, img_names_4_linear_reg, 'cmr', args)