-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathsr_inp_nonlinear.py
208 lines (193 loc) · 7.15 KB
/
sr_inp_nonlinear.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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
import argparse, os, yaml
import torch
import numpy as np
from PIL import Image
from skimage.metrics import peak_signal_noise_ratio, structural_similarity
import matplotlib.pyplot as plt
from util.img_utils import Blurkernel, clear_color, generate_tilt_map, mask_generator
from guided_diffusion.measurements import get_noise, get_operator
from guided_diffusion.unet import create_model
from ddim_sampler import *
import shutil
import lpips
def load_yaml(file_path: str) -> dict:
with open(file_path) as f:
config = yaml.load(f, Loader=yaml.FullLoader)
return config
def dmplug(model, scheduler, logdir, img='00000', eta=0, lr=1e-2, dataset='celeba',img_model_config=None,task_config=None,device='cuda'):
dtype = torch.float32
gt_img_path = './data/{}/{}.png'.format(dataset,img)
gt_img = Image.open(gt_img_path).convert("RGB")
shutil.copy(gt_img_path, os.path.join(logdir, 'gt.png'))
ref_numpy = np.array(gt_img) / 255.0
x = ref_numpy * 2 - 1
x = x.transpose(2, 0, 1)
ref_img = torch.Tensor(x).to(dtype).to(device).unsqueeze(0)
ref_img.requires_grad = False
# Prepare Operator and noise
measure_config = task_config['measurement']
operator = get_operator(device=device, **measure_config['operator'])
noiser = get_noise(**measure_config['noise'])
if measure_config['operator']['name'] == 'inpainting':
mask_gen = mask_generator(
**measure_config['mask_opt']
)
mask = mask_gen(ref_img)
mask = mask[:, 0, :, :].unsqueeze(dim=0)
# Forward measurement model (Ax + n)
y = operator.forward(ref_img, mask=mask)
y_n = noiser(y)
else:
# Forward measurement model (Ax + n)
y = operator.forward(ref_img)
y_n = noiser(y)
y_n.requires_grad = False
plt.imsave(os.path.join(logdir, 'measurement.png'), clear_color(y_n))
# DMPlug
Z = torch.randn((1, 3, img_model_config['image_size'], img_model_config['image_size']), device=device, dtype=dtype, requires_grad=True)
criterion = torch.nn.MSELoss().to(device)
params_group1 = {'params': Z, 'lr': lr}
optimizer = torch.optim.Adam([params_group1])
epochs = 5000 # SR, inpainting: 5,000, nonlinear deblurring: 10,000
psnrs = []
ssims = []
losses = []
lpipss = []
loss_fn_alex = lpips.LPIPS(net='alex').to(device)
for iterator in range(epochs):
model.eval()
optimizer.zero_grad()
for i, tt in enumerate(scheduler.timesteps):
t = (torch.ones(1) * tt).cuda()
if i == 0:
noise_pred = model(Z, t)
else:
noise_pred = model(x_t, t)
noise_pred = noise_pred[:, :3]
if i == 0:
x_t = scheduler.step(noise_pred, tt, Z, return_dict=True, use_clipped_model_output=True, eta=eta).prev_sample
else:
x_t = scheduler.step(noise_pred, tt, x_t, return_dict=True, use_clipped_model_output=True, eta=eta).prev_sample
output = torch.clamp(x_t, -1, 1)
if measure_config['operator']['name'] == 'inpainting':
loss = criterion(operator.forward(output, mask=mask), y_n)
else:
loss = criterion(operator.forward(output), y_n)
loss.backward()
optimizer.step()
losses.append(loss.item())
with torch.no_grad():
output_numpy = output.detach().cpu().squeeze().numpy()
output_numpy = (output_numpy + 1) / 2
output_numpy = np.transpose(output_numpy, (1, 2, 0))
# calculate psnr
tmp_psnr = peak_signal_noise_ratio(ref_numpy, output_numpy)
psnrs.append(tmp_psnr)
# calculate ssim
tmp_ssim = structural_similarity(ref_numpy, output_numpy, channel_axis=2, data_range=1)
ssims.append(tmp_ssim)
# calculate lpips
rec_img_torch = torch.from_numpy(output_numpy).permute(2, 0, 1).unsqueeze(0).float().to(device)
gt_img_torch = torch.from_numpy(ref_numpy).permute(2, 0, 1).unsqueeze(0).float().to(device)
rec_img_torch = rec_img_torch * 2 - 1
gt_img_torch = gt_img_torch * 2 - 1
lpips_alex = loss_fn_alex(gt_img_torch, rec_img_torch).item()
lpipss.append(lpips_alex)
if len(psnrs) == 1 or (len(psnrs) > 1 and tmp_psnr > np.max(psnrs[:-1])):
best_img = output_numpy
plt.imsave(os.path.join(logdir, "rec_img.png"), best_img)
plt.plot(np.array(losses), label='all')
plt.legend()
plt.savefig(os.path.join(logdir, 'loss.png'))
plt.close()
plt.plot(np.array(psnrs))
plt.title('Max PSNR: {}'.format(np.max(np.array(psnrs))))
plt.savefig(os.path.join(logdir, 'psnr.png'))
plt.close()
psnr_res = np.max(psnrs)
ssim_res = np.max(ssims)
lpips_res = np.min(lpipss)
print('PSNR: {}, SSIM: {}, LPIPS: {}'.format(psnr_res, ssim_res, lpips_res))
def get_parser():
parser = argparse.ArgumentParser()
parser.add_argument(
"-e",
"--eta",
type=float,
nargs="?",
help="eta for ddim sampling (0.0 yields deterministic sampling)",
default=0.0
)
parser.add_argument(
"-l",
"--logdir",
type=str,
nargs="?",
help="logdir",
default="./results"
)
parser.add_argument(
"--dataset",
type=str,
nargs="?",
help="dataset",
default="celeba"
)
parser.add_argument(
"-c",
"--custom_steps",
type=int,
nargs="?",
help="number of steps for ddim and fast sampling",
default=3
)
parser.add_argument(
"--lr",
type=float,
nargs="?",
help="lr of z",
default=0.01
)
parser.add_argument(
"--task",
type=str,
nargs="?",
help="super_resolution,inpainting,nonlinear_deblur",
default='super_resolution'
)
parser.add_argument(
"--img",
type=int,
nargs="?",
help="image id",
default=0
)
return parser
def torch_seed(seed=0):
np.random.seed(seed)
torch.manual_seed(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.cuda.manual_seed_all(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
if __name__ == "__main__":
torch_seed(123)
# Load configurations
parser = get_parser()
device = torch.device("cuda")
opt, unknown = parser.parse_known_args()
img_model_config = 'configs/model_config_{}.yaml'.format(opt.dataset)
task_config = 'configs/tasks/{}_config.yaml'.format(opt.task)
img_model_config = load_yaml(img_model_config)
model = create_model(**img_model_config)
model = model.to(device)
model.eval()
task_config = load_yaml(task_config)
# Define the DDIM scheduler
scheduler = DDIMScheduler()
scheduler.set_timesteps(opt.custom_steps)
img = str(opt.img).zfill(5)
logdir = os.path.join(opt.logdir, opt.task, opt.dataset, img)
os.makedirs(logdir,exist_ok=True)
# DMPlug
dmplug(model, scheduler, logdir, img=img, eta=opt.eta, lr=opt.lr, dataset=opt.dataset, img_model_config=img_model_config, task_config = task_config, device=device)