-
Notifications
You must be signed in to change notification settings - Fork 3
/
eval_ano.py
258 lines (192 loc) · 8.27 KB
/
eval_ano.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
#!/usr/bin/env python3
import os
import argparse
from copy import deepcopy
from functools import partial
from pathlib import Path
import accelerate
import torch
from torch import optim
from torch import multiprocessing as mp
from torch.utils import data
from tqdm.auto import tqdm
import k_diffusion as K
import matplotlib.pyplot as plt
import random
import numpy as np
GLOBAL_SEED = 42
def seed_everything(seed=42):
np.random.seed(seed)
random.seed(seed)
torch.manual_seed(seed)
global GLOBAL_SEED
GLOBAL_SEED = seed
# torch.use_deterministic_algorithms(True)
def args_parser():
p = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
p.add_argument('--config', type=str, required=False,
help='the configuration file')
p.add_argument('--batch_test', type=int, default=256,
help='the test batch size')
p.add_argument('--lr', type=float,
help='the learning rate')
p.add_argument('--n_sample', type=int,
help='evaluate sample size')
p.add_argument('--name', type=str, default='model',
help='the name of the run')
p.add_argument('--num-workers', type=int, default=6,
help='the number of data loader workers')
p.add_argument('--seed', type=int, default=42,
help='the random seed')
p.add_argument('--start-method', type=str, default='fork',
choices=['fork', 'forkserver', 'spawn'],
help='the multiprocessing start method')
return p.parse_args()
def get_datasets():
from feat_load import ClipDataset
test_set = ClipDataset(
root_dir='data',
dataset_name='shanghai',
)
return test_set
def main():
args = args_parser()
seed_everything(seed=args.seed)
mp.set_start_method(args.start_method)
torch.backends.cuda.matmul.allow_tf32 = True
config = K.config.load_config(open('configs/config_ano.json'))
model_config = config['model']
opt_config = config['optimizer']
sched_config = config['lr_sched']
ema_sched_config = config['ema_sched']
feat_size = model_config['input_size']
test_set = get_datasets()
if model_config["sampler"] == "lms":
p_sampler_fn = partial(K.sampling.sample_lms, disable=True)
elif model_config["sampler"] == "heun":
p_sampler_fn = partial(K.sampling.sample_heun, disable=True)
elif model_config["sampler"] == "dpm2":
p_sampler_fn = partial(K.sampling.sample_dpm_2, disable=True)
else:
print('unknown sampler method')
ValueError('Invalid sampler method')
accelerator = accelerate.Accelerator(gradient_accumulation_steps=1,
cpu=False)
device = accelerator.device
print(f'Process {accelerator.process_index} using device: {device}', flush=True)
gvad_model = K.models.GVADModel(
feat_size,
)
if opt_config['type'] == 'adamw':
opt = optim.AdamW(gvad_model.parameters(),
lr=opt_config['lr'] if args.lr is None else args.lr,
betas=tuple(opt_config['betas']),
eps=opt_config['eps'],
weight_decay=opt_config['weight_decay'])
elif opt_config['type'] == 'sgd':
opt = optim.SGD(gvad_model.parameters(),
lr=opt_config['lr'] if args.lr is None else args.lr,
momentum=opt_config.get('momentum', 0.),
nesterov=opt_config.get('nesterov', False),
weight_decay=opt_config.get('weight_decay', 0.))
else:
raise ValueError('Invalid optimizer type')
if sched_config['type'] == 'inverse':
sched = K.utils.InverseLR(opt,
inv_gamma=sched_config['inv_gamma'],
power=sched_config['power'],
warmup=sched_config['warmup'])
elif sched_config['type'] == 'exponential':
sched = K.utils.ExponentialLR(opt,
num_steps=sched_config['num_steps'],
decay=sched_config['decay'],
warmup=sched_config['warmup'])
else:
raise ValueError('Invalid schedule type')
assert ema_sched_config['type'] == 'inverse'
ema_sched = K.utils.EMAWarmup(power=ema_sched_config['power'],
max_value=ema_sched_config['max_value'])
if accelerator.is_main_process:
try:
print('Number of items in testset:', len(test_set))
except TypeError:
pass
test_dl = data.DataLoader(test_set, args.batch_test, shuffle=True, drop_last=False,
num_workers=args.num_workers, persistent_workers=True)
model, opt, test_dl = accelerator.prepare(gvad_model, opt, test_dl)
sigma_min = model_config['sigma_min']
sigma_max = model_config['sigma_max']
seed_noise_path = f'seed_noise_{feat_size}.pth'
if os.path.exists(seed_noise_path):
seed_noise = torch.load(seed_noise_path, map_location=device)
else:
seed_noise = torch.randn([1, feat_size], device=device)
torch.save(seed_noise, seed_noise_path)
model = K.config.make_denoiser_wrapper(config)(model)
model_ema = deepcopy(model)
ckpt_path = 'trained_model/model.pth'
if Path(ckpt_path).exists():
if accelerator.is_main_process:
print(f'Resuming from {ckpt_path}...')
ckpt = torch.load(ckpt_path, map_location='cpu')
accelerator.unwrap_model(model.gvad_model).load_state_dict(ckpt['model'])
accelerator.unwrap_model(model_ema.gvad_model).load_state_dict(ckpt['model_ema'])
opt.load_state_dict(ckpt['opt'])
sched.load_state_dict(ckpt['sched'])
ema_sched.load_state_dict(ckpt['ema_sched'])
del ckpt
else:
print('state path does not exist.')
print(ckpt_path)
exit(1)
def plot_prediction(g_dist, label, vids, idx, start_n):
g_dist = K.utils.normalize(g_dist)
uniques = np.unique(vids)
for item in uniques:
v_path = f'plots/{item}'
K.utils.mkdir(v_path)
mask = np.where(vids == item)[0]
sort_idx = np.argsort(idx[mask])
vid_dist = g_dist[mask]
vid_dist = vid_dist[sort_idx]
vid_label = label[mask]
vid_label = vid_label[sort_idx]
plt.title(f'video_{item}_{start_n}')
plt.subplot(211); plt.plot(vid_dist, label='dist', c='b')
plt.tick_params(axis='x',
which='both',
bottom=False,
top=False,
labelbottom=False)
plt.ylim(0, 1.1)
plt.legend()
plt.subplot(212); plt.plot(vid_label, label='gt', c='r')
plt.legend()
plt.ylim(0, 1.1)
plt.savefig(f'{v_path}/video_{item}_{start_n}.png')
plt.close()
@torch.no_grad()
@K.utils.eval_mode(model_ema)
def evaluate(n_start):
if accelerator.is_main_process:
tqdm.write('Evaluating...')
sigmas = K.sampling.get_sigmas_karras(10, sigma_min, sigma_max, rho=7., device=device)
sigmas = sigmas[n_start:]
sample_noise = seed_noise.to(device) * sigmas[0]
def sample_fn(x_real):
x_real = x_real.to(device)
x = sample_noise + x_real
x_0 = p_sampler_fn(model_ema.forward, x, sigmas)
g_dists = model_ema.gvad_model.loss(x_real, x_0)
return g_dists
g_dists, labels, vid, idx = K.evaluation.compute_eval_outs_aot(accelerator, sample_fn, test_dl)
vid = np.concatenate(vid)
return g_dists.cpu().numpy(), labels.cpu().numpy(), vid, idx.cpu().numpy()
for i in range(10):
g_dists, labels, vids, idx = evaluate(i)
plot_prediction(g_dists, labels, vids, idx, i)
if __name__ == '__main__':
print('Hello there!')
main()
print('Obiwan Kenobi.')