-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathsampling.py
167 lines (131 loc) · 6.58 KB
/
sampling.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
import jax
import numpy as np
import optax
import cv2
from torch.utils import data
from flax.training import train_state, checkpoints
from flax.core import freeze
from dataset.data_loader import SceneClassDataset
from model.xunet import XUNet
def cosine_beta_schedule(timesteps, s = 0.008):
"""
cosine schedule
as proposed in https://openreview.net/forum?id=-NEXDKk8gZ
"""
steps = timesteps + 1
x = np.linspace(0, timesteps, steps, dtype = np.float64)
alphas_cumprod = np.cos(((x / timesteps) + s) / (1 + s) * np.pi * 0.5) ** 2
alphas_cumprod = alphas_cumprod / alphas_cumprod[0]
betas = 1 - (alphas_cumprod[1:] / alphas_cumprod[:-1])
return np.clip(betas, 0, 0.9999)
betas = cosine_beta_schedule(1000)
alphas = 1. - betas
alphas_cumprod = np.cumprod(alphas, axis=0)
alphas_cumprod_prev = np.pad(alphas_cumprod[:-1], (1, 0), 'constant', constant_values=(1))
sqrt_alphas_cumprod = np.sqrt(alphas_cumprod)
sqrt_one_minus_alphas_cumprod = np.sqrt(1. - alphas_cumprod)
sqrt_recip_alphas_cumprod = np.sqrt(1. / alphas_cumprod)
sqrt_recipm1_alphas_cumprod = np.sqrt(1. / alphas_cumprod - 1)
posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod)
posterior_log_variance_clipped = np.log(posterior_variance.clip(min =1e-20))
posterior_mean_coef1 = betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod)
posterior_mean_coef2 = (1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod)
def predict_start_from_noise(x_t, t, noise):
return (sqrt_recip_alphas_cumprod[t] * x_t - sqrt_recipm1_alphas_cumprod[t] * noise)
def q_posterior(x_start, x_t, t):
posterior_mean = (
posterior_mean_coef1[t] * x_start +
posterior_mean_coef2[t] * x_t
)
pos_var = posterior_variance[t]
pos_log_var_clipped = posterior_log_variance_clipped[t]
return posterior_mean, pos_var, pos_log_var_clipped
img_sidelength = 64
ds = SceneClassDataset(root_dir='./cars_train_val',
max_num_instances=-1,
max_observations_per_instance=50,
img_sidelength=img_sidelength,
specific_observation_idcs=None,
samples_per_instance=1)
diffusion_model = XUNet()
batch_size = 1
def cycle(dl):
while True:
for data in dl:
yield data
def logsnr_schedule_cosine(t, *, logsnr_min=-20., logsnr_max=20.):
b = np.arctan(np.exp(-.5 * logsnr_max))
a = np.arctan(np.exp(-.5 * logsnr_min)) - b
return -2. * np.log(np.tan(a * t + b))
def create_sample_data(batch_size, img_sidelength):
sample = dict()
sample['x'] = np.random.random((batch_size, img_sidelength, img_sidelength, 3))
sample['z'] = np.random.random((batch_size, img_sidelength, img_sidelength, 3))
sample['logsnr'] = np.random.random((batch_size))
sample['R1'] = np.random.random((batch_size, 3, 3))
sample['t1'] = np.random.random((batch_size, 3))
sample['R2'] = np.random.random((batch_size, 3, 3))
sample['t2'] = np.random.random((batch_size, 3))
sample['K'] = np.random.random((batch_size, 3, 3))
sample['noise'] = np.random.random((batch_size, img_sidelength, img_sidelength, 3))
return sample
dl = cycle(data.DataLoader(ds,
batch_size = batch_size,
shuffle=True,
drop_last=True,
collate_fn=ds.collate_fn,
pin_memory=True))
sample = create_sample_data(batch_size, img_sidelength)
params = diffusion_model.init({'params' : jax.random.PRNGKey(0), 'dropout' : jax.random.PRNGKey(1)},
sample,
cond_mask=np.zeros((batch_size)), train=True)['params']
train_state = train_state.TrainState.create(apply_fn=diffusion_model.apply, params=params, tx=optax.adam(1e-3))
loaded_model_state = checkpoints.restore_checkpoint(
ckpt_dir='checkpoints', # Folder with the checkpoints
target=train_state, # (optional) matching object to rebuild state in
prefix='model0' # Checkpoint file name prefix
)
if loaded_model_state is train_state:
raise FileNotFoundError(f"Checkpoint does not exist")
params = freeze(loaded_model_state.params)
while True:
data = next(dl)[0]
def apply_model(state, data):
def t_schedule_cosine(logsnr, *, logsnr_min=-20., logsnr_max=20.):
b = np.arctan(np.exp(-.5 * logsnr_max))
a = np.arctan(np.exp(-.5 * logsnr_min)) - b
return (((np.arctan(np.exp(logsnr / -2)) - b) / a) * 1000).astype(int)
data['z'] = np.random.randn(*data['x'].shape)
data['logsnr'] = np.ones(*data['logsnr'].shape) * -20
for time_step in range(999, -1, -1):
def p_mean_variance():
output1 = state.apply_fn({'params': params}, data, cond_mask=np.ones(data['x'].shape[0]), train=False)
output2 = state.apply_fn({'params': params}, data, cond_mask=np.zeros(data['x'].shape[0]), train=False)
w = 3
output = (1 + w) * output1 - w * output2
x_recon = predict_start_from_noise(data['z'], t=time_step, noise = output)
x_recon = np.clip(x_recon, -1., 1.)
model_mean, pos_var, pos_log_var = q_posterior(x_start=x_recon, x_t=data['z'], t=time_step)
return model_mean, pos_var, pos_log_var
def p_sample():
b = data['z'].shape[0]
model_mean, _, model_log_variance = p_mean_variance()
noise = np.random.randn(*data['z'].shape)
# no noise when t == 0
nonzero_mask = np.array(1.0 - (time_step == 0)).reshape(b, *((1,) * (len(data['z']) - 1)))
return model_mean + nonzero_mask * np.exp(0.5 * model_log_variance) * noise
data['z'] = p_sample()
data['logsnr'] = logsnr_schedule_cosine(time_step / 1000.0)
cv2.imshow('output', np.array(data['z'][0] / 2.0 + 0.5))
cv2.waitKey(0)
model_input = dict()
model_input['x'] = data['x'].numpy()
model_input['z'] = data['z'].numpy()
model_input['logsnr'] = np.array(data['logsnr'])
model_input['R1'] = data['R1'].numpy()
model_input['t1'] = data['t1'].numpy()
model_input['R2'] = data['R2'].numpy()
model_input['t2'] = data['t2'].numpy()
model_input['K'] = data['K'].numpy()
model_input['noise'] = data['noise'].numpy()
apply_model(loaded_model_state, model_input)