-
Notifications
You must be signed in to change notification settings - Fork 8
/
spectrum_evolutions.py
361 lines (308 loc) · 11.5 KB
/
spectrum_evolutions.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
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
"""Analyze frequency evolution of diffusion models."""
import argparse
import pickle
from pathlib import Path
from typing import List
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from cytoolz import functoolz
from sklearn.linear_model import LogisticRegressionCV
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from tqdm import tqdm
from src.data import apply_to_imgdir, array_from_imgdir
from src.image import dct, fft, log_scale, spectral_density
from src.paper_utils import configure_matplotlib, get_figsize, get_filename
from src.visualization import plot_evolution, plot_evolution_mean
def _get_dirs(root_dir: Path, img_dirs: List[str]):
data = {}
for img_dir in sorted(root_dir.iterdir()):
if img_dirs is None or img_dir.name in img_dirs:
timesteps = int(img_dir.name.split("_")[-1])
data[timesteps] = img_dir
return {k: v for k, v in sorted(data.items())}
def spectrum_vs_steps(args):
configure_matplotlib(
rc={"figure.constrained_layout.use": False, "figure.dpi": 3000}
)
output_dir = args.output_root / spectrum_vs_steps.__name__
plot_dir = output_dir / "plots"
plot_dir.mkdir(exist_ok=True, parents=True)
reference = np.load(args.reference)
evolution, timesteps = [], []
for ts, img_dir in tqdm(_get_dirs(args.image_root, args.img_dirs).items()):
timesteps.append(ts)
evolution.append(
apply_to_imgdir(
img_dir=img_dir,
func=spectral_density,
grayscale=True,
cache_dir=output_dir / "cache",
).mean(axis=0)
)
evolution = np.array(evolution) / reference - 1
# plot evolution
if args.all_labels:
y = None
yticklabels = timesteps
else:
y = timesteps
yticklabels = None
plt.figure(figsize=get_figsize(fraction=args.fraction))
plot_evolution(data=evolution, ylabel=args.ylabel, y=y, yticklabels=yticklabels)
plt.tight_layout(pad=0.3)
plt.savefig(
plot_dir
/ get_filename(
file_format="pdf",
kind="spectrum_evolution",
data=args.image_root.name,
experiment=args.experiment,
)
)
plt.close()
# plot average
configure_matplotlib()
plt.figure(figsize=get_figsize(fraction=0.49))
plot_evolution_mean(
data=abs(evolution), x=timesteps, xlabel=args.ylabel, show_best=False
)
plt.savefig(
plot_dir
/ get_filename(
file_format="pdf",
kind="spectrum_evolution_mean",
data=args.image_root.name,
experiment=args.experiment,
overwrite=args.overwrite,
)
)
plt.close()
print(
"Timestep with minimum error:",
timesteps[np.argmin(abs(evolution).mean(axis=1))],
)
def denoising_vs_diffusion(args):
configure_matplotlib(
rc={"figure.constrained_layout.use": False, "figure.dpi": 3000}
)
output_dir = args.output_root / spectrum_vs_steps.__name__
plot_dir = output_dir / "plots"
plot_dir.mkdir(exist_ok=True, parents=True)
denoising, denoising_timesteps = [], []
for ts, img_dir in tqdm(_get_dirs(args.image_root, args.img_dirs).items()):
denoising_timesteps.append(ts)
denoising.append(
apply_to_imgdir(
img_dir=img_dir,
func=spectral_density,
grayscale=True,
cache_dir=output_dir / "cache",
overwrite=args.overwrite,
).mean(axis=0)
)
diffusion, diffusion_timesteps = [], []
for ts, img_dir in tqdm(_get_dirs(args.diffusion_root, args.img_dirs).items()):
diffusion_timesteps.append(ts)
diffusion.append(
apply_to_imgdir(
img_dir=img_dir,
func=spectral_density,
grayscale=True,
cache_dir=output_dir / "cache",
).mean(axis=0)
)
if diffusion_timesteps != denoising_timesteps:
raise ValueError("Denoising and diffusion have different timesteps.")
evolution = np.array(denoising) / np.array(diffusion) - 1
y = denoising_timesteps
yticklabels = None
plt.figure(figsize=get_figsize(fraction=args.fraction))
plot_evolution(
data=np.array(evolution), ylabel=args.ylabel, y=y, yticklabels=yticklabels
)
plt.tight_layout(pad=0.3)
plt.savefig(
plot_dir
/ get_filename(
file_format="pdf",
kind="diffusion_vs_denoising",
data=args.image_root.name,
experiment=args.experiment,
)
)
plt.close()
# plot average
configure_matplotlib()
plt.figure(figsize=get_figsize(fraction=0.49))
plot_evolution_mean(
data=abs(evolution), x=denoising_timesteps, xlabel=args.ylabel, show_best=False
)
plt.savefig(
plot_dir
/ get_filename(
file_format="pdf",
kind="diffusion_vs_denoising_mean",
data=args.image_root.name,
experiment=args.experiment,
)
)
plt.close()
print(
"Timestep with minimum error:",
denoising_timesteps[np.argmin(abs(evolution).mean(axis=1))],
)
def compute_reference(args):
output_dir = args.output_root / spectrum_vs_steps.__name__
output_dir.mkdir(parents=True, exist_ok=True)
array = array_from_imgdir(
directory=args.image_root, grayscale=True, num_samples=args.num_samples
)
ref = spectral_density(array).mean(axis=0)
np.save(output_dir / f"{args.identifier}-{args.num_samples}_samples.npy", ref)
def classification(args):
output_dir = args.output_root / spectrum_vs_steps.__name__
output_dir.mkdir(parents=True, exist_ok=True)
plot_dir = output_dir / "plots"
plot_dir.mkdir(exist_ok=True, parents=True)
transforms = {
"Pixel": lambda x: x,
"DCT": dct,
"log(DCT)": functoolz.compose_left(dct, log_scale),
"FFT": functoolz.compose_left(fft, np.abs),
"log(FFT)": functoolz.compose_left(fft, log_scale),
}
cache_file = output_dir / "cache" / f"LR_{args.image_root.name}.pickle"
if cache_file.exists():
with open(cache_file, "rb") as f:
output = pickle.load(f)
else:
real = array_from_imgdir(
directory=args.real_path,
crop_size=64,
grayscale=True,
num_samples=args.num_samples,
)
output = []
for ts, img_dir in tqdm(_get_dirs(args.image_root, args.img_dirs).items()):
fake = array_from_imgdir(
directory=img_dir, crop_size=64, grayscale=True, num_samples=512
)
x = np.concatenate([real, fake])
y = np.array([0] * len(real) + [1] * len(fake))
x_train, x_test, y_train, y_test = train_test_split(
x, y, test_size=0.2, random_state=0, stratify=y
)
for tf_name, tf_func in transforms.items():
x_train_tf = tf_func(x_train).reshape((x_train.shape[0], -1))
x_test_tf = tf_func(x_test).reshape((x_test.shape[0], -1))
scaler = StandardScaler()
x_train_tf = scaler.fit_transform(x_train_tf)
x_test_tf = scaler.transform(x_test_tf)
clf = LogisticRegressionCV(
Cs=[
1 / 1e4,
1 / 1e3,
1 / 1e2,
1 / 1e1,
1 / 1e0,
1 / 1e-1,
1 / 1e-2,
1 / 1e-3,
1 / 1e-4,
],
max_iter=1000,
n_jobs=args.num_workers,
)
clf.fit(x_train_tf, y_train)
train_score = clf.score(x_train_tf, y_train)
test_score = clf.score(x_test_tf, y_test)
output.append((ts, tf_name, train_score, test_score))
print((ts, tf_name, train_score, test_score))
with open(cache_file, "wb") as f:
pickle.dump(output, f)
df = (
pd.DataFrame(output, columns=["Step", "Transform", "Train", "Test"])
.set_index("Step")
.pivot(columns="Transform", values=["Train", "Test"])
)
# plot
configure_matplotlib()
plt.figure(figsize=get_figsize(fraction=args.fraction))
if args.all_labels:
y = range(len(df.index))
yticklabels = df.index
plt.gca().set_yticks(y, labels=yticklabels)
else:
y = df.index
plt.plot(df[("Test", "Pixel")], y, label="Pixel")
plt.plot(df[("Test", "DCT")], y, label="DCT")
plt.plot(df[("Test", "log(DCT)")], y, label="log(DCT)")
plt.plot(df[("Test", "FFT")], y, label="FFT")
plt.plot(df[("Test", "log(FFT)")], y, label="log(FFT)")
plt.ylabel(args.ylabel)
plt.xlabel("Accuracy")
plt.legend()
filename = get_filename(file_format="pdf", kind="lr_acc", data=args.image_root.name)
plt.savefig(plot_dir / filename)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument(
"--experiment",
default="default",
help="Custom experiment name to use for output files.",
)
plot = parser.add_argument_group("plot")
plot.add_argument(
"--fraction", type=float, default=1, help="Fraction of plot size."
)
plot.add_argument("--ylabel", type=str, default="$t$", help="Y-Label")
plot.add_argument("--all-labels", action="store_true", help="Plot all labels.")
data = parser.add_argument_group("data")
data.add_argument("image_root", type=Path, help="Root of image directory.")
data.add_argument("output_root", type=Path, help="Output directory.")
data.add_argument(
"--overwrite",
action="store_true",
help="Recompute instead of using existing data.",
)
data.add_argument(
"--img-dirs",
nargs="+",
type=str,
help="Subset of img-dirs in image_root to use.",
)
subparsers = parser.add_subparsers(title="subcommands", help="additional help")
p_spectrum_vs_steps = subparsers.add_parser(spectrum_vs_steps.__name__)
p_spectrum_vs_steps.set_defaults(func=spectrum_vs_steps)
p_spectrum_vs_steps.add_argument(
"reference", type=Path, help="Path to reference file."
)
p_denoising_vs_diffusion = subparsers.add_parser(denoising_vs_diffusion.__name__)
p_denoising_vs_diffusion.set_defaults(func=denoising_vs_diffusion)
p_denoising_vs_diffusion.add_argument(
"diffusion_root", type=Path, help="Path to diffusion root."
)
p_compute_reference = subparsers.add_parser(compute_reference.__name__)
p_compute_reference.set_defaults(func=compute_reference)
p_compute_reference.add_argument("identifier", help="Name of reference.")
p_compute_reference.add_argument(
"--num_samples",
type=int,
default=10000,
help="Number of samples (default: 10000).",
)
p_classification = subparsers.add_parser(classification.__name__)
p_classification.set_defaults(func=classification)
p_classification.add_argument("real_path", type=Path, help="Path to real images.")
p_classification.add_argument(
"--num_samples", type=int, default=512, help="Number of samples (default: 512)."
)
p_classification.add_argument(
"--num_workers", type=int, default=4, help="Number of workers (default: 4)."
)
args = parser.parse_args()
args.func(args)
if __name__ == "__main__":
parse_args()