-
Notifications
You must be signed in to change notification settings - Fork 2
/
scope.py
150 lines (130 loc) · 5.81 KB
/
scope.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
# ----------------------------------------------#
# Pro : SCOPE
# File : reprojection.py
# Date : 2023/4/17
# Author : Qing Wu
# Email : wuqing@shanghaitech.edu.cn
# ----------------------------------------------#
import SimpleITK as sitk
import numpy as np
import torch
import dataset
import tinycudann as tcnn
import commentjson as json
from torch.utils import data
from torch.optim import lr_scheduler
def train(config_path):
# config
# ------------------------------------
with open(config_path) as config_file:
config = json.load(config_file)
sv_sino_in_path = config["file"]["sv_sino_in_path"]
dv_sino_out_path = config["file"]["dv_sino_out_path"]
model_path = config["file"]["model_path"]
num_sv, num_dv, L = config["file"]["num_sv"], config["file"]["num_dv"], config["file"]["L"]
lr = config["train"]["lr"]
epoch = config["train"]["epoch"]
gpu = config["train"]["gpu"]
summary_epoch = config["train"]["summary_epoch"]
sample_N = config["train"]["sample_N"]
batch_size = config["train"]["batch_size"]
# data
# ------------------------------------
train_loader = data.DataLoader(
dataset=dataset.TrainData(sin_path=sv_sino_in_path, theta=num_sv, sample_N=sample_N),
batch_size=batch_size,
shuffle=True
)
# model & optimizer
# ------------------------------------
DEVICE = torch.device('cuda:{}'.format(str(gpu) if torch.cuda.is_available() else 'cpu'))
l1_loss_function = torch.nn.L1Loss() # L1 Loss
SCOPE = tcnn.NetworkWithInputEncoding(n_input_dims=2, n_output_dims=1,
encoding_config=config["encoding"],
network_config=config["network"]).to(DEVICE)
optimizer = torch.optim.Adam(params=SCOPE.parameters(), lr=lr)
scheduler = lr_scheduler.StepLR(optimizer, step_size=500, gamma=0.5)
# train
# ------------------------------------
for e in range(epoch):
SCOPE.train()
loss_train = 0
for i, (ray_sample, projection_l_sample) in enumerate(train_loader):
# the sampled rays and the corresponding projections
ray_sample = ray_sample.to(DEVICE).float().view(-1, 2) # (N, sample_N, L, 2)
projection_l_sample = projection_l_sample.to(DEVICE).float() # (N, sample_N)
# forward
pre_intensity = SCOPE(ray_sample).view(batch_size, sample_N, L, 1) # (N, sample_N, L, 1)
projection_l_sample_pre = torch.sum(pre_intensity, dim=2) # (N, sample_N, 1, 1)
# reshape
projection_l_sample_pre = projection_l_sample_pre.squeeze(-1).squeeze(-1) # (N, sample_N)
# compute loss
loss = l1_loss_function(projection_l_sample_pre,
projection_l_sample.to(projection_l_sample_pre.dtype))
# backward
optimizer.zero_grad()
loss.backward()
optimizer.step()
# record and print loss
loss_train += loss.item()
scheduler.step()
print('{}, (TRAIN0) Epoch[{}/{}], Lr:{}, Loss:{:.6f}'.
format(num_sv, e + 1, epoch, scheduler.get_last_lr()[0], loss_train/len(train_loader)))
if (e + 1) % summary_epoch == 0:
torch.save(SCOPE.state_dict(), model_path)
def reprojection(config_path):
# config
# ------------------------------------
with open(config_path) as config_file:
config = json.load(config_file)
sv_sino_in_path = config["file"]["sv_sino_in_path"]
dv_sino_out_path = config["file"]["dv_sino_out_path"]
model_path = config["file"]["model_path"]
num_sv, num_dv, L = config["file"]["num_sv"], config["file"]["num_dv"], config["file"]["L"]
scale = int(num_dv/num_sv)
batch_size = config["train"]["batch_size"]
gpu = config["train"]["gpu"]
# dataloader
# ------------------------------------
test_loader = data.DataLoader(
dataset=dataset.TestData(theta=num_dv, L=L),
batch_size=batch_size,
shuffle=False
)
# model
# ------------------------------------
DEVICE = torch.device('cuda:{}'.format(str(gpu) if torch.cuda.is_available() else 'cpu'))
SCOPE = tcnn.NetworkWithInputEncoding(n_input_dims=2, n_output_dims=1,
encoding_config=config["encoding"],
network_config=config["network"]).to(DEVICE)
SCOPE.load_state_dict(torch.load(model_path))
# reprojetion
# ------------------------------------
sin_pre = np.zeros(shape=(num_dv, L))
with torch.no_grad():
SCOPE.eval()
for i, (ray_sample) in enumerate(test_loader):
print(i, len(test_loader))
# all the parallel rays from each view
ray_sample = ray_sample.to(DEVICE).float().view(-1, 2) # (N, L, L, 2)
# forward
pre_intensity = SCOPE(ray_sample).view(-1, L, L, 1) # (N, L, L, 1)
# projection i.e, Equ. 2
projection_l_sample_pre = torch.sum(pre_intensity, dim=2) # (N, L, 1, 1)
# reshape and store
projection_l_sample_pre = projection_l_sample_pre.squeeze(-1).squeeze(-1) # (N, L)
temp = projection_l_sample_pre.cpu().detach().float().numpy()
if i == 0:
sin_pre = temp
else:
sin_pre = np.concatenate((sin_pre, temp), axis=0)
# data consistency
sin_original = sitk.GetArrayFromImage(sitk.ReadImage(sv_sino_in_path))
k = 0
for i in range(len(sin_pre)):
if i % scale == 0:
sin_pre[i, :] = sin_original[k, :]
k = k + 1
# write dense-view sinogram and model
sin_pre = sitk.GetImageFromArray(sin_pre)
sitk.WriteImage(sin_pre, '{}/{}_sino_pre.nii'.format(dv_sino_out_path, num_dv))