-
Notifications
You must be signed in to change notification settings - Fork 6
/
limited_ct_sample.py
183 lines (163 loc) · 7.22 KB
/
limited_ct_sample.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
171
172
173
174
175
176
177
178
179
180
181
182
183
################################################################################
# Copyright 2022-2023 Lawrence Livermore National Security, LLC and other
# DOLCE project developers. See the LICENSE file for details.
# SPDX-License-Identifier: MIT
#
# DOLCE: A Model-Based Probabilistic Diffusion Framework for Limited-Angle CT Reconstruction
################################################################################
"""
Generate a large batch of samples from a DOLCE model.
"""
import warnings
warnings.filterwarnings("ignore", category=DeprecationWarning)
import os
import torch
import argparse
import numpy as np
from functools import partial
import torch.distributed as dist
from guided_diffusion import dist_util, logger
from guided_diffusion.image_datasets import dataset2run
from guided_diffusion.script_util import (
condtion_model_and_diffusion_defaults,
condtion_create_model_and_diffusion,
args_to_dict,
add_dict_to_argparser,
)
from dataFidelities.CTClass import set_CTClass
def main():
args = create_argparser().parse_args()
dist_util.setup_dist()
# set random seed
torch.manual_seed(args.seed)
np.random.seed(args.seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(args.seed)
device, gpu_index, device_name = dist_util.dev()
logger.configure(data_set=args.data_dir, num_angles=args.num_angs)
#############################
logger.log("creating testing dataset...")
data = dataset2run(
data_dir=args.data_dir,
batch_size=args.batch_size,
image_size=args.image_size,
angle_range=args.num_angs,
deterministic=args.deterministic
)
data_len = data.len_data()
args.num_samples = args.num_samples if args.num_samples<=data_len else data_len
logger.log("creating denoising model...")
model, diffusion = condtion_create_model_and_diffusion(
**args_to_dict(args, condtion_model_and_diffusion_defaults().keys())
)
model.load_state_dict(
dist_util.load_state_dict(args.model_path, map_location="cpu")
)
model.to(device)
if args.use_fp16:
model.convert_to_fp16()
model.eval()
logger.log("creating data model...")
dObj = set_CTClass(device, gpu_index, use_cuda=True,
batch_size=args.batch_size, use_static=False,
param_fn=args.param_fn, rank=dist.get_rank(), logger=logger)
run_sampler = partial(diffusion.sample_loop, model_kwargs_data=dObj)
logger.log("Successfully loaded conditional model. ")
logger.log("creating samples...")
all_images_data, all_images_gt, all_images_rls = [], [], []
for gt, model_kwargs in data.load_data():
gt = gt.to(device)
model_kwargs = {k: v.to(device) for k, v in model_kwargs.items()}
sample_data, fbp, rls = run_sampler(
model,
gt, #only used for generating sine online
eta=args.eta,
shape=gt.shape,
samper=args.sampler,
prox_solver=args.prox_solver,
model_kwargs=model_kwargs,
clip_denoised=args.clip_denoised,
)
sample_data=torch.mean(sample_data,0,keepdim=True)
sample_data = sample_data.permute(0, 2, 3, 1)
gt = gt[0,...].unsqueeze(0).contiguous()
rls = rls[0,...].unsqueeze(0).contiguous()
sample_data = sample_data.contiguous()
all_gt = [torch.zeros_like(gt) for _ in range(dist.get_world_size())]
all_rls = [torch.zeros_like(rls) for _ in range(dist.get_world_size())]
all_samples_data = [torch.zeros_like(sample_data) for _ in range(dist.get_world_size())]
dist.all_gather(all_gt, gt) # gather not supported with NCCL
dist.all_gather(all_rls, rls) # gather not supported with NCCL
dist.all_gather(all_samples_data, sample_data) # gather not supported with NCCL
for sample_data in all_samples_data:
all_images_data.append(sample_data.cpu().numpy())
for gt in all_gt:
all_images_gt.append(gt.cpu().numpy())
for rls in all_rls:
all_images_rls.append(rls.cpu().numpy())
logger.log(f"created {len(all_images_data) * args.batch_size} samples")
if len(all_images_data) * args.batch_size >= args.num_samples:
break
arr_data = np.concatenate(all_images_data, axis=0)
arr_gt = np.concatenate(all_images_gt, axis=0)
arr_rls = np.concatenate(all_images_rls, axis=0)
arr_data = arr_data.transpose([1,2,3,0])[...,:args.num_samples]
arr_gt = arr_gt.transpose([2,3,1,0])[...,:args.num_samples]
arr_rls = arr_rls.transpose([2,3,1,0])[...,:args.num_samples]
logger.log(f"used {args.num_samples} samples")
if dist.get_rank() == 0:
shape_str = "x".join([str(x) for x in arr_data.shape])
out_path = os.path.join(logger.get_dir(), f"samples_{shape_str}.npz")
logger.log(f"saving to {out_path}")
np.savez(out_path, arr_gt, arr_data, arr_rls)
logger.log("calculating ssim/psnr")
psnr_all_data, ssim_all_data = [], []
for ii in range(arr_gt.shape[-1]):
arr_gdt_each = arr_gt[...,ii].squeeze()
arr_rls_each = arr_rls[...,ii].squeeze()
arr_each = arr_data[...,ii].squeeze()
psne_each = dObj.eval_psnr(arr_gdt_each, arr_each, data_range=arr_gdt_each.max() - arr_gdt_each.min())
ssim_each = dObj.eval_ssim(arr_gdt_each, arr_each, data_range=arr_gdt_each.max() - arr_gdt_each.min())
psnr_all_data.append(psne_each)
ssim_all_data.append(ssim_each)
if args.save_fig:
save_fig_pth = os.path.join(logger.get_dir(), f"samples_{ii}_ang={args.num_angs}.png")
arr_all = np.concatenate(
[dObj.np_normalize(arr_gdt_each),
dObj.np_normalize(arr_rls_each),
dObj.np_normalize(arr_each)],
axis=1,)
dObj.imsave(255*arr_all, save_fig_pth)
psnr_all_data, ssim_all_data = np.array(psnr_all_data), np.array(ssim_all_data)
# if args.save_fig:
# dObj.save_video(logger.get_dir(), logger.get_dir())
logger.log(f"PSNR_MEAN_DATA: {psnr_all_data.mean()}, PSNR_MAX_DATA: {psnr_all_data.max()}, PSNR_MIN_DATA: {psnr_all_data.min()}")
logger.log(f"SSIM_MEAN_DATA: {ssim_all_data.mean()}, SSIM_MAX_DATA: {ssim_all_data.max()}, SSIM_MIN_DATA: {ssim_all_data.min()}")
dist.barrier()
logger.log("sampling complete")
def create_argparser():
defaults = dict(
sampler='ddpm', #"ddpm" or "ddim"
prox_solver='apgm', #"apgm" or "cgrad"
use_condtion='rls', #"rls" or "fbp"
clip_denoised=True,
num_samples=10000,
batch_size=1, #bs = 1 works the best
data_dir="",
model_path="",
param_fn="",
sub_id="",
num_angs=60, #60, 90, 120
eta=1., #used in ddim
seed=12345,
weighted_condition=False,
save_fig=True,
root_path=None,
deterministic=False,
)
defaults.update(condtion_model_and_diffusion_defaults())
parser = argparse.ArgumentParser()
add_dict_to_argparser(parser, defaults)
return parser
if __name__ == "__main__":
main()