-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_demo.py
486 lines (368 loc) · 16.5 KB
/
train_demo.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
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
# -*- coding: utf-8 -*-
"""
Created on Thu Mar 12 12:28:02 2020
@author: Kitt Miller
"""
import torch
import numpy as np
from torchvision import transforms, datasets, models
import matplotlib.pyplot as plt
import torch.nn as nn
import matplotlib
from torch.autograd import Variable
import torch.nn.functional as F
from tqdm import tqdm
import argparse
import json
import tarfile
import os
import boto3
import sagemaker
from sagemaker.pytorch import PyTorchModel
from sagemaker.predictor import RealTimePredictor, json_serializer, json_deserializer
## utils
def imshow(image_tensor, mean, std, title=None):
"""
Imshow for normalized Tensors.
Useful to visualize data from data loader
"""
image = image_tensor.numpy().transpose((1, 2, 0))
image = std * image + mean
image = np.clip(image, 0, 1)
plt.imshow(image)
if title is not None:
plt.title(title)
plt.pause(0.001) # pause a bit so that plots are updated
def accuracy(output, target):
_, predicted = torch.max(output.data, 1)
total = target.size(0)
correct = (predicted == target).sum().item()
accuracy = correct/total
return accuracy
class Tracker:
def __init__(self):
self.data = {}
def track(self, name, *monitors):
l = Tracker.ListStorage(monitors)
self.data.setdefault(name, []).append(l)
return l
def to_dict(self):
return {k: list(map(list, v)) for k, v in self.data.items()}
class ListStorage:
def __init__(self, monitors=[]):
self.data = []
self.monitors = monitors
for monitor in self.monitors:
setattr(self, monitor.name, monitor)
def append(self, item):
for monitor in self.monitors:
monitor.update(item)
self.data.append(item)
def __iter__(self):
return iter(self.data)
class MeanMonitor:
name = 'mean'
def __init__(self):
self.n = 0
self.total = 0
def update(self, value):
self.total += value
self.n += 1
@property
def value(self):
return self.total / self.n
class MovingMeanMonitor:
name = 'mean'
def __init__(self, momentum=0.9):
self.momentum = momentum
self.first = True
self.value = None
def update(self, value):
if self.first:
self.value = value
self.first = False
else:
m = self.momentum
self.value = m * self.value + (1 - m) * value
## Plotting
def plotter():
path_adapt = "log_adaptation.pth"
# Load logs
log_adapt = torch.load(path_adapt)
# compute mean for each epoch
adapt = {
'classification_loss': torch.FloatTensor(log_adapt['classification_loss']).mean(dim=1).numpy(),
'coral_loss': torch.FloatTensor(log_adapt['CORAL_loss']).mean(dim=1).numpy(),
'source_accuracy': torch.FloatTensor(log_adapt['source_accuracy']).mean(dim=1).numpy(),
'target_accuracy': torch.FloatTensor(log_adapt['target_accuracy']).mean(dim=1).numpy()
}
# Add the first 0 value
adapt['target_accuracy'] = np.insert(adapt['target_accuracy'], 0, 0)
adapt['source_accuracy'] = np.insert(adapt['source_accuracy'], 0, 0)
axes = plt.gca()
axes.set_ylim([0, 1.1])
l1, = plt.plot(adapt['target_accuracy'], label="test acc. w/ coral loss", marker='*')
l3, = plt.plot(adapt['source_accuracy'], label="training acc. w/ coral loss", marker='^')
plt.legend(handles=[l1, l3], loc=4)
fig_acc = plt.gcf()
plt.show()
plt.figure()
fig_acc.savefig('accuracies.pdf', dpi=1000)
# Classification loss and CORAL loss for training w/ CORAL loss
axes = plt.gca()
axes.set_ylim([0, 0.5])
l5, = plt.plot(adapt['classification_loss'], label="classification loss", marker='.')
l6, = plt.plot(adapt['coral_loss'], label="coral loss", marker='*')
plt.legend(handles=[l5, l6], loc=1)
fig_acc = plt.gcf()
plt.show()
plt.figure()
fig_acc.savefig('losses_adapt.pdf', dpi=1000)
## Model
class Net(nn.Module):
def __init__(self, num_classes, pretrained=True):
super(Net, self).__init__()
# check https://github.com/pytorch/vision/blob/master/torchvision/models/alexnet.py
self.model = models.densenet161(pretrained=pretrained, num_classes=num_classes)
# if we want to feed 448x448 images
# self.model.avgpool = nn.AdaptiveAvgPool2d(1)
# In case we want to apply the loss to any other layer than the last
# we need a forward hook on that layer
# def save_features_layer_x(module, input, output):
# self.layer_x = output
# This is a forward hook. Is executed each time forward is executed
# self.model.layer4.register_forward_hook(save_features_layer_x)
def forward(self, x):
out = self.model(x)
return out # , self.layer_x
## Dataloading
def get_loader(name_dataset, batch_size, mean_std, train=True):
# Computed with compute_mean_std.py
transform = {
'train': transforms.Compose(
[transforms.Resize([256, 256]),
transforms.RandomCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(mean=mean_std[name_dataset]['mean'],
std=mean_std[name_dataset]['std'])]),
'test': transforms.Compose(
[transforms.Resize([224, 224]),
transforms.ToTensor(),
transforms.Normalize(mean=mean_std[name_dataset]['mean'],
std=mean_std[name_dataset]['std'])])
}
dataset = datasets.ImageFolder(root='./data/%s' % name_dataset,
transform=transform['train' if train else 'test'])
dataset_loader = torch.utils.data.DataLoader(dataset,
batch_size=batch_size, shuffle=train,
num_workers=4)
return dataset_loader
## Coral
def coral(source, target):
d = source.size(1) # dim vector
source_c = compute_covariance(source)
target_c = compute_covariance(target)
loss = torch.sum(torch.mul((source_c - target_c), (source_c - target_c)))
loss = loss / (4 * d * d)
return loss
def compute_covariance(input_data):
"""
Compute Covariance matrix of the input data
"""
n = input_data.size(0) # batch_size
# Check if using gpu or cpu
if input_data.is_cuda:
device = torch.device('cuda')
else:
device = torch.device('cpu')
id_row = torch.ones(n).resize(1, n).to(device=device)
sum_column = torch.mm(id_row, input_data)
mean_column = torch.div(sum_column, n)
term_mul_2 = torch.mm(mean_column.t(), mean_column)
d_t_d = torch.mm(input_data.t(), input_data)
c = torch.add(d_t_d, (-1 * term_mul_2)) * 1 / (n - 1)
return c
## Mean + Stdev
def compute_mean_std(path_dataset):
"""
Compute mean and standard deviation of an image dataset.
Acknowledgment : http://forums.fast.ai/t/image-normalization-in-pytorch/7534
"""
transform = transforms.Compose([
transforms.Resize(224),
transforms.ToTensor()
])
dataset = datasets.ImageFolder(root=path_dataset,
transform=transform)
# Choose a large batch size to better approximate. Optimally load the dataset entirely on memory.
data_loader = torch.utils.data.DataLoader(dataset, batch_size=4096, shuffle=False, num_workers=4)
pop_mean = []
pop_std = []
for i, data in enumerate(data_loader, 0):
# shape (batch_size, 3, height, width)
numpy_image = data[0].numpy()
# shape (3,) -> 3 channels
batch_mean = np.mean(numpy_image, axis=(0, 2, 3))
batch_std = np.std(numpy_image, axis=(0, 2, 3))
pop_mean.append(batch_mean)
pop_std.append(batch_std)
# shape (num_iterations, 3) -> (mean across 0th axis) -> shape (3,)
pop_mean = np.array(pop_mean).mean(axis=0)
pop_std = np.array(pop_std).mean(axis=0)
values = {
'mean': pop_mean,
'std': pop_std
}
return values
def dataset_stats():
mean_std = {}
for dataset in ['labeled', 'unlabeled', 'test']:
# Construct path
dataset_path = './data/%s' % dataset
values = compute_mean_std(dataset_path)
# Add values to dict
mean_std[dataset] = values
return mean_std
## Training
def train(model, optimizer, source_loader, target_loader, tracker, args, epoch=0):
model.train()
tracker_class, tracker_params = tracker.MovingMeanMonitor, {'momentum': 0.99}
# Trackers to monitor classification and CORAL loss
classification_loss_tracker = tracker.track('classification_loss', tracker_class(**tracker_params))
coral_loss_tracker = tracker.track('CORAL_loss', tracker_class(**tracker_params))
min_n_batches = min(len(source_loader), len(target_loader))
tq = tqdm(range(min_n_batches), desc='{} E{:03d}'.format('Training + Adaptation', epoch), ncols=0)
for _ in tq:
source_data, source_label = next(iter(source_loader))
target_data, _ = next(iter(target_loader)) # Unsupervised Domain Adaptation
source_data, source_label = Variable(source_data.to(device=args.device)), Variable(source_label.to(device=args.device))
target_data = Variable(target_data.to(device=args.device))
optimizer.zero_grad()
out_source = model(source_data)
out_target = model(target_data)
classification_loss = F.cross_entropy(out_source, source_label)
# This is where the magic happens
coral_loss = coral(out_source, out_target)
composite_loss = classification_loss + args.lambda_coral * coral_loss
composite_loss.backward()
optimizer.step()
classification_loss_tracker.append(classification_loss.item())
coral_loss_tracker.append(coral_loss.item())
fmt = '{:.4f}'.format
tq.set_postfix(classification_loss=fmt(classification_loss_tracker.mean.value),
coral_loss=fmt(coral_loss_tracker.mean.value))
def evaluate(model, data_loader, dataset_name, tracker, args, epoch=0):
model.eval()
tracker_class, tracker_params = tracker.MeanMonitor, {}
acc_tracker = tracker.track('{}_accuracy'.format(dataset_name), tracker_class(**tracker_params))
loader = tqdm(data_loader, desc='{} E{:03d}'.format('Evaluating on %s' % dataset_name, epoch), ncols=0)
accuracies = []
with torch.no_grad():
for target_data, target_label in loader:
target_data = Variable(target_data.to(device=args.device))
target_label = Variable(target_label.to(device=args.device))
output = model(target_data)
accuracies.append(accuracy(output, target_label))
acc_tracker.append(sum(accuracies)/len(accuracies))
fmt = '{:.4f}'.format
loader.set_postfix(accuracy=fmt(acc_tracker.mean.value))
def main():
# Paper: In the training phase, we set the batch size to 128,
# base learning rate to 10−3, weight decay to 5×10−4, and momentum to 0.9
parser = argparse.ArgumentParser(description='Train - Evaluate DeepCORAL model')
parser.add_argument('--disable_cuda', action='store_true',
help='Disable CUDA')
parser.add_argument('--epochs', type=int, default=25,
help='Number of total epochs to run')
parser.add_argument('--batch_size', type=int, default=32,
help='Batch size')
parser.add_argument('--lr', type=float, default=0.0005,
help='Learning Rate')
parser.add_argument('--decay', default=5e-4,
help='Decay of the learning rate')
parser.add_argument('--momentum', default=0.9,
help="Optimizer's momentum")
parser.add_argument('--lambda_coral', type=float, default=0.8,
help="Weight that trades off the adaptation with "
"classification accuracy on the source domain")
parser.add_argument('--source', default='labeled',
help="Source Domain (dataset)")
parser.add_argument('--target', default='unlabeled',
help="Target Domain (dataset)")
parser.add_argument('--test', default='test',
help="Test/Validation Data (dataset)")
parser.add_argument('--naming', default='unnamed',
help="Model save name")
parser.add_argument('--bucket', default='domaintestkitt',
help="S3 bucket for output file")
args = parser.parse_args()
args.device = None
if not args.disable_cuda and torch.cuda.is_available():
args.device = torch.device('cuda')
print('CUDA enabled')
else:
args.device = torch.device('cpu')
print('CUDA disabled or not available')
stats = dataset_stats()
source_train_loader = get_loader(name_dataset=args.source, batch_size=args.batch_size, mean_std =stats, train=True)
target_train_loader = get_loader(name_dataset=args.target, batch_size=args.batch_size, mean_std =stats, train=True)
source_evaluate_loader = get_loader(name_dataset=args.source, batch_size=args.batch_size, mean_std =stats, train=False)
target_evaluate_loader = get_loader(name_dataset=args.test, batch_size=args.batch_size, mean_std =stats, train=False)
n_classes = len(source_train_loader.dataset.classes)
# ~ Paper : "We initialized the other layers with the parameters pre-trained on ImageNet"
model = models.densenet169(pretrained=True)
# ~ Paper : The dimension of last fully connected layer (fc8) was set to the number of categories (31)
model.classifier = nn.Linear(1664, n_classes)
# ~ Paper : and initialized with N(0, 0.005)
torch.nn.init.normal_(model.classifier.weight, mean=0, std=5e-3)
# Initialize bias to small constant number (http://cs231n.github.io/neural-networks-2/#init)
model.classifier.bias.data.fill_(0.01)
model = model.to(device=args.device)
# ~ Paper : "The learning rate of fc8 is set to 10 times the other layers as it was training from scratch."
optimizer = torch.optim.SGD([
{'params': model.features.parameters()},
{'params': model.classifier.parameters(), 'lr': 5 * args.lr}
], lr=args.lr, momentum=args.momentum) # if not specified, the default lr is used
tracker = Tracker()
for i in range(args.epochs):
train(model, optimizer, source_train_loader, target_train_loader, tracker, args, i)
evaluate(model, source_evaluate_loader, 'source', tracker, args, i)
evaluate(model, target_evaluate_loader, 'target', tracker, args, i)
# Save logged classification loss, coral loss, source accuracy, target accuracy
torch.save(tracker.to_dict(), "log_adaptation.pth")
torch.save(model, str(str(args.naming) +'_model.pt'))
with open('model.pth', 'wb') as f:
torch.save(model.state_dict(), f)
for i in stats.keys():
for j in ['mean', 'std']:
stats[i][j] = np.ndarray.tolist(stats[i][j])
model_data = {
'num_classes': n_classes,
'classes':source_train_loader.dataset.classes,
'data_stats': stats,
}
with open('model_data.json', 'w') as fp:
json.dump(model_data, fp)
tar = tarfile.open("output.tar.gz", "w:gz")
for name in ["model_data.json", "log_adaptation.pth", str(str(args.naming) +'_model.pt'), 'model.pth']:
tar.add(name)
tar.close()
role = *<REPLACE WITH YOUR SAGEMAKER ENABLE ROLE ARN>*
s3client = boto3.client('s3')
with open('output.tar.gz', "rb") as f:
s3client.upload_fileobj(f, args.bucket, 'output.tar.gz')
class JSONPredictor(RealTimePredictor):
def __init__(self, endpoint_name, sagemaker_session):
super(JSONPredictor, self).__init__(endpoint_name, sagemaker_session, json_serializer, json_deserializer)
model = PyTorchModel(model_data="s3://{}/output.tar.gz".format(args.bucket),
role=role,
framework_version='1.4.0',
entry_point='serve_demo.py',
predictor_cls=JSONPredictor)
predictor = model.deploy(initial_instance_count=1, instance_type='ml.t2.medium')
with open("endpoint.txt", "w") as write_file:
write_file.write(predictor.endpoint)
if __name__ == '__main__':
main()
plotter()