forked from facebookresearch/videoalignment
-
Notifications
You must be signed in to change notification settings - Fork 1
/
rmac_features.py
397 lines (340 loc) · 12.7 KB
/
rmac_features.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
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
# Copyright (c) Facebook, Inc. and its affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
#
import argparse
import os
import pickle as pkl
import tempfile
from subprocess import call
import numpy as np
import torch
import torch.nn as nn
from torch.utils.data import DataLoader, Dataset
from torchvision import transforms
from torchvision.datasets.folder import default_loader
from torchvision.models import resnet34
DESCRIPTION = """
"RMAC features extractor"
"""
class VideoDataset(Dataset):
"""
A simple Dataset class for loading images inside a folder.
"""
def __init__(self, video_path, transforms, fps, image_area):
self.video_path = video_path
self.transforms = transforms
self.root_dir = tempfile.TemporaryDirectory()
self.root = self.root_dir.name
FNULL = open(os.devnull, "w")
call(
[
"ffmpeg",
"-i",
self.video_path,
"-filter:v",
"fps=%f" % fps,
os.path.join(self.root + "/%07d.jpg"),
],
stderr=FNULL,
)
self.images = sorted(os.listdir(self.root))
self.images = [os.path.join(self.root, i) for i in self.images]
self.transforms.transforms[0].size = get_scale(self.images[0], image_area)
def __getitem__(self, item):
path = self.images[item]
img = default_loader(path)
if self.transforms:
img = self.transforms(img)
return img
def __len__(self):
return len(self.images)
class PCA(object):
"""
Fits and applies PCA whitening
"""
def __init__(self, n_components, device):
self.n_components = n_components
self.mean = None
self.DVt = None
self.device = device
def fit(self, X):
mean = X.mean(axis=0)
X -= mean
self.mean = torch.from_numpy(mean).view(1, -1).to(self.device)
Xcov = np.dot(X.T, X)
d, V = np.linalg.eigh(Xcov)
idx = np.argsort(d)[::-1][: self.n_components]
d = d[idx]
V = V[:, idx]
D = np.diag(1.0 / np.sqrt(d + 10e-8))
self.DVt = torch.from_numpy(np.dot(D, V.T)).to(self.device)
def apply(self, X):
X -= self.mean.expand_as(X)
num = torch.mm(self.DVt, X.transpose(0, 1)).transpose(0, 1)
out = num / torch.sqrt(torch.sum(num ** 2, dim=1, keepdim=True))
return out
def get_rmac_region_coordinates(H, W, L):
# Almost verbatim from Tolias et al Matlab implementation.
# Could be heavily pythonized, but really not worth it...
# Desired overlap of neighboring regions
ovr = 0.4
# Possible regions for the long dimension
steps = np.array((2, 3, 4, 5, 6, 7), dtype=np.float32)
w = np.minimum(H, W)
b = (np.maximum(H, W) - w) / (steps - 1)
# steps(idx) regions for long dimension. The +1 comes from Matlab
# 1-indexing...
idx = np.argmin(np.abs(((w ** 2 - w * b) / w ** 2) - ovr)) + 1
# Region overplus per dimension
Wd = 0
Hd = 0
if H < W:
Wd = idx
elif H > W:
Hd = idx
regions_xywh = []
for l in range(1, L + 1):
wl = np.floor(2 * w / (l + 1))
wl2 = np.floor(wl / 2 - 1)
# Center coordinates
if l + Wd - 1 > 0:
b = (W - wl) / (l + Wd - 1)
else:
b = 0
cenW = np.floor(wl2 + b * np.arange(l - 1 + Wd + 1)) - wl2
# Center coordinates
if l + Hd - 1 > 0:
b = (H - wl) / (l + Hd - 1)
else:
b = 0
cenH = np.floor(wl2 + b * np.arange(l - 1 + Hd + 1)) - wl2
for i_ in cenH:
for j_ in cenW:
regions_xywh.append([j_, i_, wl, wl])
# Round the regions. Careful with the borders!
for i in range(len(regions_xywh)):
for j in range(4):
regions_xywh[i][j] = int(round(regions_xywh[i][j]))
if regions_xywh[i][0] + regions_xywh[i][2] > W:
regions_xywh[i][0] -= (regions_xywh[i][0] + regions_xywh[i][2]) - W
if regions_xywh[i][1] + regions_xywh[i][3] > H:
regions_xywh[i][1] -= (regions_xywh[i][1] + regions_xywh[i][3]) - H
return np.array(regions_xywh).astype(np.float32)
class ResNet34_ith(nn.Module):
"""
ResNet-34 cut at the i-th layer
"""
def __init__(self, n):
super(ResNet34_ith, self).__init__()
self.n = n
self.model = resnet34(pretrained=True)
self.model.eval()
def forward(self, x):
x = self.model.conv1(x)
x = self.model.bn1(x)
x = self.model.relu(x)
counter = 1
if counter == self.n:
return x
x = self.model.maxpool(x)
for block in [
self.model.layer1,
self.model.layer2,
self.model.layer3,
self.model.layer4,
]:
for layer in block._modules.values():
residual = x
x = layer.conv1(x)
x = layer.bn1(x)
x = layer.relu(x)
counter += 1
if counter == self.n:
break
x = layer.conv2(x)
x = layer.bn2(x)
if layer.downsample is not None:
residual = layer.downsample(residual)
x += residual
x = layer.relu(x)
counter += 1
if counter == self.n:
break
return x
def get_scale(image_path, n_pixels):
image = default_loader(image_path)
factor = np.sqrt((image.width * image.height) / n_pixels)
return int(np.rint(min(image.width, image.height) / factor))
def get_rmac_descriptors(cnn, args, video_path, aggregated=True, pca=None):
image_transforms = transforms.Compose(
[
transforms.Resize(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
]
)
print("Create video dataset")
dataset = VideoDataset(video_path, image_transforms, args.fps, args.image_area)
print("Extract features")
dataloader = DataLoader(dataset, batch_size=args.b_s, num_workers=args.workers)
devices_ids = list(range(torch.cuda.device_count()))
if devices_ids:
print(f"Use devices: {devices_ids}")
parallel_cnn = nn.DataParallel(cnn, device_ids=devices_ids)
else:
parallel_cnn = cnn
all_rmac_descriptors = []
dataloader_iter = iter(dataloader)
with torch.no_grad():
for it, x_images in enumerate(dataloader_iter):
images = x_images.float().to(args.device)
features = parallel_cnn(images)
nc = features.size()[1]
args.pca_dimensions = nc
rmac_regions = get_rmac_region_coordinates(
features.size()[2], features.size()[3], args.rmac_levels
).astype(np.int)
rmac_descriptors = []
for region in rmac_regions:
desc = torch.max(
torch.max(
features[
:,
:,
region[1] : (region[3] + region[1]),
region[0] : (region[2] + region[0]),
],
2,
keepdim=True,
)[0],
3,
keepdim=True,
)[0]
rmac_descriptors.append(desc.view(-1, 1, nc))
rmac_descriptors = torch.cat(rmac_descriptors, 1)
nr = rmac_descriptors.size()[1]
# L2-norm
rmac_descriptors = rmac_descriptors.view(-1, nc)
rmac_descriptors = rmac_descriptors / torch.sqrt(
torch.sum(rmac_descriptors ** 2, dim=1, keepdim=True)
)
if args.regional_descriptors:
rmac_descriptors = rmac_descriptors.view(-1, nr, nc)
if not aggregated:
rmac_descriptors = rmac_descriptors.cpu().data.numpy()
all_rmac_descriptors.append(rmac_descriptors)
else:
# PCA whitening
rmac_descriptors = pca.apply(rmac_descriptors)
# Sum aggregation and L2-normalization
rmac_descriptors = torch.sum(
rmac_descriptors.view(-1, nr, args.pca_dimensions), 1
)
rmac_descriptors = rmac_descriptors / torch.sqrt(
torch.sum(rmac_descriptors ** 2, dim=1, keepdim=True)
)
rmac_descriptors = rmac_descriptors.detach().cpu().numpy()
all_rmac_descriptors.append(rmac_descriptors)
return np.concatenate(all_rmac_descriptors)
def train_pca(cnn, args):
out_dir = args.out_folder
sum_pooling = args.sum_before_pca
directory = os.path.join(out_dir, "tmp_%d/" % args.resnet_level)
if not os.path.isdir(directory):
os.mkdir(directory)
outfile = directory + os.path.basename(args.dataset_folder) + ".pkl"
if os.path.exists(outfile):
return
descriptors = []
for dir, subdirs, files in os.walk(args.dataset_folder):
if files and not subdirs:
for file in files:
file = os.path.join(dir, file)
print("Computing descriptors for %s" % file)
try:
desc = get_rmac_descriptors(cnn, args, file, aggregated=False)
if sum_pooling:
desc = np.sum(desc, 1)
desc = desc / np.sqrt(
np.sum(desc ** 2, axis=-1, keepdims=True) + 10e-8
)
descriptors.append(desc)
except:
print("Unable to process %s" % file)
descriptors = np.concatenate(descriptors)
pca = PCA(args.pca_dimensions, args.device)
pca.fit(descriptors)
if sum_pooling:
pca_prefix = args.pca_files_prefix + "_sum"
else:
pca_prefix = args.pca_files_prefix
torch.save(
pca.DVt,
os.path.join(
args.dataset_folder, pca_prefix + "_Dvt_resnet34_%d.t7" % args.resnet_level
),
)
torch.save(
pca.mean,
os.path.join(
args.dataset_folder, pca_prefix + "_mean_resnet34_%d.t7" % args.resnet_level
),
)
print("Done.")
def compute_features(cnn, args):
pca = PCA(args.pca_dimensions, args.device)
pca.DVt = torch.load(
os.path.join(
args.dataset_folder,
args.pca_files_prefix + "_Dvt_resnet34_%d.t7" % args.resnet_level,
),
map_location="cpu",
).to(args.device)
pca.mean = torch.load(
os.path.join(
args.dataset_folder,
args.pca_files_prefix + "_mean_resnet34_%d.t7" % args.resnet_level,
),
map_location="cpu",
).to(args.device)
in_filename = args.dataset_folder
print("Computing descriptors for %s" % in_filename)
descriptors = get_rmac_descriptors(cnn, args, in_filename, pca=pca)
filename = os.path.basename(in_filename)
filename, _ = os.path.splitext(filename)
out_folder = os.path.join(args.out_folder, f"rmac_resnet34_{args.resnet_level}")
fname_desc = os.path.join(out_folder, filename + ".pkl")
os.makedirs(out_folder, exist_ok=True)
pkl.dump(descriptors, open(fname_desc, "wb"), -1)
def main(args):
cnn = ResNet34_ith(args.resnet_level)
cnn = cnn.to(args.device)
if args.train_pca:
train_pca(cnn, args)
else:
compute_features(cnn, args)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description=DESCRIPTION)
parser.add_argument(
"dataset_folder",
help="path to dataset (to compute the PCA), or to a single video file",
)
parser.add_argument("pca_files_prefix", help="Prefix for files for PCA")
parser.add_argument("out_folder", help="Output folder for descriptors files")
parser.add_argument("--train_pca", action="store_true")
parser.add_argument("--regional_descriptors", action="store_true")
parser.add_argument("--image_area", default=12000, type=int)
parser.add_argument("--resnet_level", default=30, type=int)
parser.add_argument("--rmac_levels", default=3, type=int)
parser.add_argument("--pca_dimensions", default=512, type=int)
parser.add_argument("--workers", default=16, type=int)
parser.add_argument("--b_s", default=200, type=int)
parser.add_argument("--fps", default=15, type=float)
parser.add_argument("--sum_before_pca", action="store_true")
args = parser.parse_args()
args.device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Running with args={args}")
main(args)