-
Notifications
You must be signed in to change notification settings - Fork 3
/
tpdm_utils.py
149 lines (118 loc) · 5.42 KB
/
tpdm_utils.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
import torch
import torchmetrics as tm
import numpy as np
import os
import argparse
from typing import Union
from models import utils as mutils
from models.ema import ExponentialMovingAverage
from utils import restore_checkpoint
from sde_lib import VESDE
def get_tpdm_sde(config):
sigmas = mutils.get_sigmas(config)
if config.training.sde.lower() == 'vesde':
sde = VESDE(sigma_min=config.model.sigma_min, sigma_max=config.model.sigma_max, N=config.model.num_scales)
else:
raise NotImplementedError("TPDM is only implemented for VESDE")
return sde, sigmas
def get_tpdm_models(config, ckpt_pri_path, ckpt_aux_path):
score_model_pri = mutils.create_model(config)
score_model_aux = mutils.create_model(config)
ema_pri = ExponentialMovingAverage(score_model_pri.parameters(), decay=config.model.ema_rate)
ema_aux = ExponentialMovingAverage(score_model_aux.parameters(), decay=config.model.ema_rate)
state_pri = dict(step=0, model=score_model_pri, ema=ema_pri)
state_aux = dict(step=0, model=score_model_aux, ema=ema_aux)
state_pri = restore_checkpoint(ckpt_pri_path, state_pri, config.device, skip_optimizer=True)
state_aux = restore_checkpoint(ckpt_aux_path, state_aux, config.device, skip_optimizer=True)
ema_pri.copy_to(score_model_pri.parameters())
ema_aux.copy_to(score_model_aux.parameters())
return score_model_pri, score_model_aux
def load_tpdm_label_data(path):
fname_list = [str(fname.name) for fname in path.glob("*.npy")]
fname_list = sorted(fname_list, key=lambda x: float(x.split(".")[0]))
fname_list = [(f"{i:03d}.npy", True) if f"{i:03d}.npy" in fname_list else (f"{i:03d}.npy", False)
for i in range(256)]
assert len(fname_list) == 256
print("Loading all data ...")
all_img = []
for fname in fname_list:
fname, isfile = fname
if isfile:
img = np.load(os.path.join(path, fname)).squeeze()
if np.issubdtype(img.dtype.type, np.integer):
img = img.astype(np.float32) / 255
elif np.issubdtype(img.dtype.type, np.floating):
img = img.astype(np.float32)
else:
raise NotImplementedError(f"Image type {img.dtype.type} is not supported")
img = torch.from_numpy(img)
else:
img = torch.zeros((256, 256), dtype=torch.float32)
h, w = img.shape
assert (h, w) == (256, 256)
img = img.view(1, 1, h, w)
all_img.append(img)
all_img = torch.cat(all_img, dim=0)
print(f"Data loaded shape: {all_img.shape}, min: {all_img.min()}, max: {all_img.max()}")
print("Note: please check the data range is in about [0, 1]")
return all_img, fname_list
def eval_recon_result(volume_recon, volume_label, plane, clip=True):
if clip:
volume_recon = torch.clip(volume_recon, 0.0, 1.0)
volume_label = torch.clip(volume_label, 0.0, 1.0)
if plane == "coronal":
pass
elif plane == "sagittal":
volume_recon = volume_recon.permute((2, 1, 0, 3))
volume_label = volume_label.permute((2, 1, 0, 3))
pass
elif plane == "axial":
volume_recon = volume_recon.permute((3, 1, 0, 2))
volume_label = volume_label.permute((3, 1, 0, 2))
pass
else:
raise ValueError(f"Unknown plane {plane}")
psnr = tm.functional.peak_signal_noise_ratio(volume_recon, volume_label, data_range=1.0 if clip else None)
psnr = psnr.item()
ssim = tm.functional.structural_similarity_index_measure(volume_recon, volume_label, data_range=1.0 if clip else None)
ssim = ssim.item()
return psnr, ssim
def print_and_save_eval_result(volume_recon, volume_label, save_root, clip=True):
# evaluate result
psnr_c, ssim_c = eval_recon_result(volume_recon, volume_label, "coronal", clip=clip)
psnr_s, ssim_s = eval_recon_result(volume_recon, volume_label, "sagittal", clip=clip)
psnr_a, ssim_a = eval_recon_result(volume_recon, volume_label, "axial", clip=clip)
# print result
print("\n<Evaluation result>")
print(f"PSNR (coronal) : {psnr_c:.4f}, SSIM (coronal) : {ssim_c:.4f}")
print(f"PSNR (sagittal) : {psnr_s:.4f}, SSIM (sagittal) : {ssim_s:.4f}")
print(f"PSNR (axial) : {psnr_a:.4f}, SSIM (axial) : {ssim_a:.4f}")
# save result
with open(save_root / 'result.txt', 'w') as f:
f.write(f"PSNR (coronal) : {psnr_c:.4f}, SSIM (coronal) : {ssim_c:.4f}\n")
f.write(f"PSNR (sagittal) : {psnr_s:.4f}, SSIM (sagittal) : {ssim_s:.4f}\n")
f.write(f"PSNR (axial) : {psnr_a:.4f}, SSIM (axial) : {ssim_a:.4f}\n")
def int_or_float(value):
try:
return int(value)
except ValueError:
try:
return float(value)
except ValueError:
raise argparse.ArgumentTypeError(f"Invalid number: {value}")
def check_K(K: Union[int, float]):
if isinstance(K, int):
if K < 2:
raise ValueError(f"K should be greater than 1 when K is integer, but got {K}")
elif isinstance(K, float):
if K <= 0.0:
raise ValueError(f"K should be greater than 0 when K is float, but got {K}")
else:
assert False, f"Unexpected type {type(K)} for K"
def is_primary_tern(i: int, K: Union[int, float]) -> bool:
if isinstance(K, int):
return i % K != K - 1
elif isinstance(K, float):
return (torch.rand(1) > 1/K).item()
else:
assert False, f"Unexpected type {type(K)} for K"