-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
265 lines (210 loc) · 8.19 KB
/
train.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
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
import numpy as np
import torch
import math
import copy
from os import listdir
from os.path import isfile, join
from pathlib import Path
from random import random
from functools import partial
from collections import namedtuple, OrderedDict
from multiprocessing import cpu_count
from PIL import Image
import logging
import logging.config
import torch
from torch import nn, einsum
import torch.nn.functional as F
from torch.nn.utils import clip_grad_norm_
from torch.utils.data import Dataset, DataLoader
from torchvision import utils
from torch.optim import Adam
from torch.optim.lr_scheduler import StepLR
from einops import rearrange, reduce
from einops.layers.torch import Rearrange
from tqdm.auto import tqdm
from ema_pytorch import EMA
from guided_diffusion.ddpm.kernel_diffusion import KernelDiffusion
from guided_diffusion.ddpm.gaussian_diffusion import default, extract, identity, exists, cycle, convert_image_to_fn
from guided_diffusion.ddpm.gaussian_diffusion import num_to_groups, has_int_squareroot
from models.deep_weiner.deblur import DEBLUR
from models.unet_kernel_y import KernelUNet
from util.dataloaders import BlurDataset
import wandb
def move_to_(data, device):
if isinstance(data,list):
new_list = []
for data_obj in data:
new_list.append(data_obj.to(device))
return new_list
else:
data.to(device)
return data
class MyDataParallel(nn.DataParallel):
def __getattr__(self, name):
try:
return super().__getattr__(name)
except AttributeError:
return getattr(self.module, name)
def check_nan(t):
return torch.isnan(t).any()
# trainer class
class Trainer(object):
def __init__(
self,
diffusion_model,
train_datasets,
val_datasets,
kernel_list,
device,
train_batch_size = 16,
gradient_accumulate_every = 2,
augment_horizontal_flip = True,
train_lr = 1e-4,
loss_x_coeff = 0.0,
loss_reblur_coeff = 0.0,
train_num_steps = 1000000,
ema_update_every = 10,
ema_decay = 0.995,
adam_betas = (0.9, 0.99),
save_and_sample_every = 5000,
num_samples = 4,
results_folder = 'results/',
amp = False,
fp16 = False,
split_batches = True,
convert_image_to = None,
):
super().__init__()
self.model = diffusion_model
assert has_int_squareroot(num_samples), 'number of samples must have an integer square root'
self.num_samples = num_samples
self.session = wandb.init(project="Kernel-Diff Training")
self.save_and_sample_every = save_and_sample_every
self.device = device
self.batch_size = train_batch_size
self.gradient_accumulate_every = gradient_accumulate_every
self.train_num_steps = train_num_steps
self.image_size = diffusion_model.image_size
# dataset and dataloader
num_workers = cpu_count()
self.ds = BlurDataset(folder_list=train_datasets, kernel_list = kernel_list)
dl = DataLoader(self.ds, batch_size = train_batch_size, shuffle = True, pin_memory = True, num_workers = num_workers)
self.dl = cycle(dl)
self.ds_val = BlurDataset(folder_list=val_datasets, kernel_list = kernel_list)
dl_val = DataLoader(self.ds_val, batch_size = num_samples, shuffle = True, pin_memory = True, num_workers = num_workers)
self.dl_val = cycle(dl_val)
# optimizer
self.reduce_lr = train_num_steps//2
self.opt = Adam(diffusion_model.model.parameters(), lr = train_lr, betas = adam_betas)
self.scheduler = StepLR(self.opt, step_size = 200000, gamma=0.5)
self.ema = EMA(self.model, beta = ema_decay, update_every = ema_update_every)
# # For loss visualization
# self.plotter = VisdomLinePlotter(env_name='Diffusion for Kernel Estimation')
self.log_step = 10
self.results_folder = Path(results_folder)
self.results_folder.mkdir(exist_ok = True)
# step counter state
self.step = 0
def save(self, milestone):
data = {
'step': self.step,
'model': self.ema.ema_model.state_dict(),
'opt': self.opt.state_dict(),
}
torch.save(data, str(self.results_folder / f'model-aided-latest.pt'))
def train(self):
device = self.device
with tqdm(initial = self.step, total = self.train_num_steps) as pbar:
self.model.train()
while self.step < self.train_num_steps:
total_loss = 0.
running_losses = np.asarray([0.0])
for _ in range(self.gradient_accumulate_every):
data = next(self.dl)
data = move_to_(data, device)
loss_k = self.model(data)
loss = loss_k
loss = loss / self.gradient_accumulate_every
total_loss += loss.item()
# running_losses += np.asarray([loss_k.item()])
loss.backward()
clip_grad_norm_(self.model.parameters(), 1.0)
pbar.set_description(f'loss: {total_loss:.4f}')
self.session.log({"loss": loss.item()})
self.opt.step()
self.opt.zero_grad()
self.ema.to(device)
self.ema.update()
self.scheduler.step()
if self.step % self.save_and_sample_every == 0:
self.ema.ema_model.eval()
milestone = self.step // self.save_and_sample_every
with torch.no_grad():
batches = num_to_groups(self.num_samples, self.batch_size)
for idx in range(len(batches)):
true_kernels_list = []
est_kernels_list = []
est_gt_list = []
blur_list = []
gt_list = []
data = move_to_(next(self.dl_val), device)
y, kernel = data[1], data[2]
x_hat, kernel_hat, k_list = self.ema.ema_model.sample(y)
true_kernels_list.append(kernel)
est_kernels_list.append(kernel_hat)
est_gt_list.append(x_hat)
blur_list.append(y)
gt_list.append(data[0])
true_kernels = torch.cat(true_kernels_list, dim = 0)
blur = torch.cat(blur_list, dim = 0)
gt = torch.cat(gt_list, dim = 0)
est_kernels = torch.cat(est_kernels_list, dim = 0)
est_gt = torch.cat(est_gt_list, dim = 0)
mse_error = -10*np.log10(torch.mean((gt-est_gt)**2).item())
print('Epoch: ', milestone)
print('Validation PSNR: {:.3f}'.format(mse_error))
utils.save_image(true_kernels, str(self.results_folder / f'sample-true-{milestone}.png'), nrow = int(math.sqrt(self.num_samples)))
utils.save_image(blur, str(self.results_folder / f'blur-{milestone}.png'), nrow = int(math.sqrt(self.num_samples)))
utils.save_image(gt, str(self.results_folder / f'gt-{milestone}.png'), nrow = int(math.sqrt(self.num_samples)))
utils.save_image(est_kernels, str(self.results_folder / f'sample-est-{milestone}.png'), nrow = int(math.sqrt(self.num_samples)))
utils.save_image(est_gt, str(self.results_folder / f'sample-est-gt-{milestone}.png'), nrow = int(math.sqrt(self.num_samples)))
self.save(milestone)
self.model.train()
self.step += 1
pbar.update(1)
if __name__ == "__main__":
import argparse
from os import path, listdir
from os.path import join, isdir
parser = argparse.ArgumentParser(description='training parameters')
parser.add_argument('--batch_size', type=int, default=4)
parser.add_argument('--save_every', type=int, default=10000)
parser.add_argument('--training_dir', type=str, default='data/LSIDR/')
parser.add_argument('--dwdn_model', type=str, default='model_zoo/model_DWDN.pt')
parser.add_argument('--kernel_file', type=str, default='data/kernel_list.npy')
args = parser.parse_args()
# Load LSIDR directory and leave one out as validation set
lsidr = listdir(args.training_dir)
list_of_dirs = []
for directory in lsidr:
if isdir(join(args.training_dir, directory)):
list_of_dirs.append(join(args.training_dir, directory))
train_datasets = list_of_dirs[0:-1]
val_datasets = [list_of_dirs[-1]]
# Generate list of kernels for training
kernel_list = [ ]
from motionblur.motionblur import Kernel
print('Generating motion blur kernels ...')
for idx in tqdm(range(6000)):
intensity = np.clip( 0.2 + 0.8*np.random.uniform(),0,0.99)
kernel = Kernel(size=(64,64), intensity=intensity).kernelMatrix
kernel_list.append(kernel)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = KernelUNet().to(device)
dwdn = DEBLUR().to(device)
dwdn.load_state_dict(torch.load(args.dwdn_model))
diffusion = KernelDiffusion(model, dwdn, image_size = 256, train_loss = 'l2').to(device)
trainer = Trainer(diffusion, train_datasets, val_datasets, kernel_list,
device = device, train_batch_size = args.batch_size, num_samples = args.batch_size, train_lr = 1e-5, save_and_sample_every = args.save_every)
trainer.train()