File size: 5,214 Bytes
28e6f98
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from diffusion_pytorch import GaussianDiffusion, Trainer, Model
from Fid import calculate_fid_given_samples
import torchvision
import os
import errno
import shutil
import argparse


def create_folder(path):
    try:
        os.mkdir(path)
    except OSError as exc:
        if exc.errno != errno.EEXIST:
            raise
        pass


def del_folder(path):
    try:
        shutil.rmtree(path)
    except OSError as exc:
        pass


create = 0

if create:
    trainset = torchvision.datasets.CIFAR10(
            root='./data', train=False, download=True)
    root = './root_cifar10_test/'
    del_folder(root)
    create_folder(root)

    for i in range(10):
        lable_root = root + str(i) + '/'
        create_folder(lable_root)

    for idx in range(len(trainset)):
        img, label = trainset[idx]
        print(idx)
        img.save(root + str(label) + '/' + str(idx) + '.png')


parser = argparse.ArgumentParser()
parser.add_argument('--time_steps', default=50, type=int)
parser.add_argument('--sample_steps', default=None, type=int)
parser.add_argument('--kernel_std', default=0.1, type=float)
parser.add_argument('--save_folder', default='progression_cifar', type=str)
parser.add_argument('--load_path', default='/cmlscratch/eborgnia/cold_diffusion/paper_defading_random_1/model.pt', type=str)
parser.add_argument('--data_path', default='./root_cifar10_test/', type=str)
parser.add_argument('--test_type', default='test_paper_showing_diffusion_images_diff', type=str)
parser.add_argument('--fade_routine', default='Random_Incremental', type=str)
parser.add_argument('--sampling_routine', default='x0_step_down', type=str)
parser.add_argument('--remove_time_embed', action="store_true")
parser.add_argument('--discrete', action="store_true")
parser.add_argument('--residual', action="store_true")

args = parser.parse_args()
print(args)

img_path=None
if 'train' in args.test_type:
    img_path = args.data_path
elif 'test' in args.test_type:
    img_path = args.data_path

print("Img Path is ", img_path)



image_channels = 1

if model_name == "unet":
    model = Model(resolution=args.image_size,
                  in_channels=1,
                  out_ch=1,
                  ch=128,
                  ch_mult=(1, 2, 2, 2),
                  num_res_blocks=2,
                  attn_resolutions=(16,),
                  dropout=0.1).cuda()

elif model_name == "twounet":

    model = TwoBranchNewModel(resolution=args.image_size,
                  in_channels=1,
                  out_ch=1,
                  ch=128,
                  ch_mult=(1, 2, 2, 2),
                  num_res_blocks=2,
                  attn_resolutions=(16,),
                  dropout=0.1).cuda()

elif model_name == "twobranch":
    downsample = [4, 4, 4]
    disc_channels = 64
    disc_layers = 3
    discriminator_iter_start = 10000
    disc_loss_type = "hinge"
    image_gan_weight = 1.0
    video_gan_weight = 1.0
    l1_weight = 4.0
    gan_feat_weight = 4.0
    perceptual_weight = 4.0
    i3d_feat = False
    restart_thres = 1.0
    no_random_restart = False
    norm_type = "group"
    padding_type = "replicate"
    num_groups = 32

    base_num_every_group = 2
    num_features = 64
    act = "PReLU"
    num_channels = 1

    model = TwoBranchModel(
        image_channels,
        disc_channels, disc_layers, disc_loss_type,
        gan_feat_weight, image_gan_weight,
        discriminator_iter_start,
        perceptual_weight, l1_weight,
        num_features, act, base_num_every_group, num_channels
    ).cuda()


diffusion = GaussianDiffusion(
    diffusion_type,
    model,
    image_size=args.image_size,    # Used to be 32
    channels=image_channels,
    device_of_kernel='cuda',
    timesteps=args.time_steps,
    loss_type=args.loss_type,  #$'l1',
    kernel_std=args.kernel_std,
    fade_routine=args.fade_routine,
    sampling_routine=args.sampling_routine,
    discrete=args.discrete
).cuda()


trainer = Trainer(
    diffusion,
    img_path,
    image_size = 32,
    train_batch_size = 32,
    train_lr = 2e-5,
    train_num_steps = 700000,         # total training steps
    gradient_accumulate_every = 2,    # gradient accumulation steps
    ema_decay = 0.995,                # exponential moving average decay
    fp16 = False,                       # turn on mixed precision training with apex
    results_folder = args.save_folder,
    load_path = args.load_path
)




if args.test_type == 'train_data':
    trainer.test_from_data('train', s_times=args.sample_steps)

elif args.test_type == 'test_data':
    trainer.test_from_data('test', s_times=args.sample_steps)

elif args.test_type == 'mixup_train_data':
    trainer.test_with_mixup('train')

elif args.test_type == 'mixup_test_data':
    trainer.test_with_mixup('test')

elif args.test_type == 'test_random':
    trainer.test_from_random('random')

elif args.test_type == 'test_fid_distance_decrease_from_manifold':
    trainer.fid_distance_decrease_from_manifold(calculate_fid_given_samples, start=0, end=None)

elif args.test_type == 'test_paper_invert_section_images':
    trainer.paper_invert_section_images()

elif args.test_type == 'test_paper_showing_diffusion_images_diff':
    trainer.paper_showing_diffusion_images()