-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathClassifier.py
389 lines (334 loc) · 18.9 KB
/
Classifier.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
import os
from enum import Enum
from types import SimpleNamespace
import torch
from pytorch_lightning.loggers import TensorBoardLogger
from sklearn.metrics import confusion_matrix
from torch.utils.data import DataLoader, WeightedRandomSampler
from torchvision.transforms import (CenterCrop,
Compose,
Normalize,
RandomHorizontalFlip,
RandomResizedCrop,
Resize,
ToTensor, RandomVerticalFlip)
from pytorch_lightning import Trainer
from pytorch_lightning.callbacks import EarlyStopping, ModelCheckpoint
from Classifiers.finetune_resnet50_classifier import ResNet50LightningModule
from Classifiers.finetune_vit_classifier import ViTLightningModule
from ood_and_id_dataset import OODAndIDDataset
from utils import create_ood_id_dataset
from plot_results import plot_confusion_matrix
from resnet_pytorch_small_images.train import create_and_train_model
from resnet_pytorch_small_images.utils import get_network, best_acc_weights, most_recent_folder
from utils import eval_model, ResizeLargestAndPad
from transformers import ViTImageProcessor
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
DATE_FORMAT = '%A_%d_%B_%Y_%Hh_%Mm_%Ss'
class DatasetType(Enum):
IN_DISTRIBUTION = "id"
OUT_OF_DISTRIBUTION = "ood"
NONE = ""
class Classifier:
def __init__(self, output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names):
self.in_dist_train_dataloader = None
self.in_dist_test_dataloader = None
self.current_run_name = current_run_name
self.classifier_cfg = classifier_cfg
self.data_source = data_source
self.ood_class_names = ood_class_names
self.train_output_dir = os.path.join(output_dir, classifier_cfg.train_output_dir, classifier_cfg.type, current_run_name)
self.checkpoint_path = os.path.join(self.train_output_dir, self.classifier_cfg.checkpoint_path)
self.test_output_dir = os.path.join(output_dir, classifier_cfg.test_output_dir, classifier_cfg.type, current_run_name)
os.makedirs(self.test_output_dir, exist_ok=True)
os.makedirs(self.train_output_dir, exist_ok=True)
self.logger = logger
def preperae_id_ood_datasets(self, data_source, ood_class_names):
# create OOD and ID datasets folders according to Anomaly Detection stage results
create_ood_id_dataset(src_root=os.path.join(data_source, "train"),
target_root=os.path.join(data_source, "train_ood_id_split"),
ood_classes=ood_class_names)
create_ood_id_dataset(src_root=os.path.join(data_source, "val"),
target_root=os.path.join(data_source, "val_ood_id_split"),
ood_classes=ood_class_names)
create_ood_id_dataset(src_root=os.path.join(data_source, "test"),
target_root=os.path.join(data_source, "test_ood_id_split"),
ood_classes=ood_class_names)
def evaluate_classifier(self, best_model):
self.logger.info("Evaluate trained model...")
if self.classifier_cfg.evaluate:
all_preds, all_labels = eval_model(dataloader=self.in_dist_test_dataloader, model=best_model, cache_dir=self.test_output_dir)
res_confusion_matrix = confusion_matrix(all_labels, all_preds)
plot_confusion_matrix(confusion_matrix=res_confusion_matrix,
classes=self.in_dist_train_dataloader.dataset.classes,
normalize=True,
output_dir=self.test_output_dir)
plot_confusion_matrix(confusion_matrix=res_confusion_matrix,
classes=self.in_dist_train_dataloader.dataset.classes,
normalize=False,
output_dir=self.test_output_dir)
def initiate_trainer(self):
pass
def train(self):
pass
def collate_fn(examples):
pixel_values = torch.stack([example[0] for example in examples])
labels = torch.tensor([example[1] for example in examples])
return {"pixel_values": pixel_values, "labels": labels}
def create_dataloaders(train_transforms, val_transforms, data_paths, dataset_type: DatasetType,
train_batch_size=100, val_batch_size=100,
dataloader_num_workers=4, ood_classes_names=None, use_weighted_sampler=False):
if ood_classes_names is None:
ood_classes_names = []
train_ds = OODAndIDDataset(root_dir=os.path.join(data_paths["train"]) if dataset_type is DatasetType.NONE
else os.path.join(data_paths["train"], f"{dataset_type.value}_dataset"),
dataset_type="train",
transform=train_transforms,
ood_classes_names=ood_classes_names)
val_ds = OODAndIDDataset(
root_dir=os.path.join(data_paths["val"]) if dataset_type is DatasetType.NONE else
os.path.join(data_paths["val"], f"{dataset_type.value}_dataset"),
dataset_type="val",
transform=val_transforms,
ood_classes_names=ood_classes_names)
test_ds = OODAndIDDataset(
root_dir=os.path.join(data_paths["test"]) if dataset_type is DatasetType.NONE else
os.path.join(data_paths["test"], f"{dataset_type.value}_dataset"),
dataset_type="test",
transform=val_transforms,
ood_classes_names=ood_classes_names)
train_sampler = WeightedRandomSampler(train_ds.weights, len(train_ds), replacement=True) if use_weighted_sampler else None
train_dataloader = DataLoader(train_ds,
batch_size=train_batch_size,
num_workers=dataloader_num_workers, sampler=train_sampler, shuffle=True)
val_sampler = WeightedRandomSampler(val_ds.weights, len(val_ds), replacement=True) if use_weighted_sampler else None
val_dataloader = DataLoader(val_ds,
batch_size=val_batch_size,
num_workers=dataloader_num_workers, sampler=val_sampler)
# test_sampler = WeightedRandomSampler(test_ds.weights, len(test_ds), replacement=True) if use_weighted_sampler else None
test_dataloader = DataLoader(test_ds,
batch_size=val_batch_size,
num_workers=dataloader_num_workers)
return train_dataloader, val_dataloader, test_dataloader
class ResNetClassifier(Classifier):
def __init__(self, output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names):
super().__init__(output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names)
self.resize = None
def get_resnet_transforms(self):
mean = [0.485, 0.456, 0.406]
std = [0.229, 0.224, 0.225]
self.train_transforms = Compose([
ResizeLargestAndPad(self.resize),
RandomHorizontalFlip(),
RandomVerticalFlip(),
ToTensor(),
Normalize(mean, std)
])
self.val_transforms = Compose([
ResizeLargestAndPad(self.resize),
ToTensor(),
Normalize(mean, std)
])
class ResNet50Classifier(ResNetClassifier):
def __init__(self, output_dir, classifier_cfg, logger, current_run_name):
super().__init__(output_dir, classifier_cfg, logger, current_run_name)
self.resize=224
def train(self):
self.initiate_trainer()
if self.classifier_cfg.retrain:
self.logger.info(f"Starting to train the ResNet50 classifier\n")
early_stop_callback = EarlyStopping(
monitor='validation_loss',
patience=3,
strict=True,
verbose=True,
mode='min'
)
checkpoint_callback = ModelCheckpoint(
monitor='validation_loss',
mode='min',
save_top_k=1,
dirpath=self.checkpoint_path,
filename='best_model',
)
logger = TensorBoardLogger(save_dir=os.path.join(self.train_output_dir, "tb_logs"), name=self.current_run_name)
trainer = Trainer(num_nodes=1, max_epochs=self.classifier_cfg.max_epoch,
callbacks=[early_stop_callback, checkpoint_callback], logger=logger)
ckpt_path = os.path.join(self.checkpoint_path, "best_model.ckpt") if self.classifier_cfg.resume else None
trainer.fit(self.model, ckpt_path=ckpt_path)
"""Finally, let's test the trained model on the test set:"""
results = trainer.test()
print(results)
def initiate_trainer(self):
self.logger.info(f"Initializing classifier\n")
self.preperae_id_ood_datasets(self.data_source, self.ood_class_names)
self.get_resnet_transforms()
self.in_dist_train_dataloader, self.in_dist_val_dataloader, self.in_dist_test_dataloader = create_dataloaders(
train_transforms=self.train_transforms, val_transforms=self.val_transforms,
data_paths={"train": os.path.join(self.data_source, "train_ood_id_split"),
"val": os.path.join(self.data_source, "val_ood_id_split"),
"test": os.path.join(self.data_source, "test_ood_id_split"),
}, dataset_type=DatasetType.IN_DISTRIBUTION, val_batch_size=self.classifier_cfg.val_batch_size,
train_batch_size=self.classifier_cfg.train_batch_size,
dataloader_num_workers=self.classifier_cfg.dataloader_num_workers,
use_weighted_sampler=self.classifier_cfg.weighted_sampler)
batch = next(iter(self.in_dist_train_dataloader))
for k, v in batch.items():
if isinstance(v, torch.Tensor):
print(k, v.shape)
assert batch['pixel_values'].shape == (self.classifier_cfg.train_batch_size, 3, self.resize, self.resize)
assert batch['labels'].shape == (self.classifier_cfg.train_batch_size,)
self.id2label = self.in_dist_train_dataloader.dataset.labels_to_classes_names
self.label2id = self.in_dist_train_dataloader.dataset.classes_names_to_labels
self.model = ResNet50LightningModule(train_dataloader=self.in_dist_train_dataloader,
val_dataloader=self.in_dist_val_dataloader,
test_dataloader=self.in_dist_test_dataloader,
loss_class_weights=self.classifier_cfg.loss_class_weights,
max_epochs=self.classifier_cfg.max_epoch,
num_labels=len(self.id2label))
self.model = self.model.to(device)
def get_fine_tuned_model(self):
model = ResNet50LightningModule.load_from_checkpoint(
os.path.join(self.checkpoint_path, f"best_model.ckpt"),
train_dataloader=self.in_dist_train_dataloader,
val_dataloader=self.in_dist_val_dataloader,
test_dataloader=self.in_dist_test_dataloader,
id2label=self.id2label,
label2id=self.label2id,
num_labels=len(self.label2id)
)
return model
class ResNet18Classifier(ResNetClassifier):
def __init__(self, output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names):
super().__init__(output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names)
self.resize=32
def initiate_trainer(self):
self.logger.info(f"Initializing classifier\n")
self.preperae_id_ood_datasets(self.data_source, self.ood_class_names)
self.get_resnet_transforms()
self.in_dist_train_dataloader, self.in_dist_val_dataloader, self.in_dist_test_dataloader = create_dataloaders(
train_transforms=self.train_transforms, val_transforms=self.val_transforms,
data_paths={"train": os.path.join(self.data_source, "train_ood_id_split"),
"val": os.path.join(self.data_source, "val_ood_id_split"),
"test": os.path.join(self.data_source, "test_ood_id_split"),
}, dataset_type=DatasetType.IN_DISTRIBUTION, val_batch_size=self.classifier_cfg.val_batch_size,
train_batch_size=self.classifier_cfg.train_batch_size,
dataloader_num_workers=self.classifier_cfg.dataloader_num_workers,
use_weighted_sampler=self.classifier_cfg.weighted_sampler)
batch = next(iter(self.in_dist_train_dataloader))
for k, v in batch.items():
if isinstance(v, torch.Tensor):
print(k, v.shape)
assert batch['pixel_values'].shape == (self.classifier_cfg.train_batch_size, 3, self.resize, self.resize)
assert batch['labels'].shape == (self.classifier_cfg.train_batch_size,)
def train(self):
self.initiate_trainer()
if self.classifier_cfg.retrain:
self.logger.info("Start training...")
create_and_train_model(train_dataloader=self.in_dist_train_dataloader,
val_dataloader=self.in_dist_val_dataloader,
checkpoint=self.checkpoint_path,
log_dir=self.train_output_dir,
logger=self.logger,
net_type=self.classifier_cfg.type,
resume=self.classifier_cfg.resume,
max_epoch=self.classifier_cfg.max_epoch,
milestones=self.classifier_cfg.milestones,
loss_class_weights=self.classifier_cfg.loss_class_weights)
def get_fine_tuned_model(self):
args = SimpleNamespace()
args.net = self.classifier_cfg.type
args.gpu = True
args.num_classes = len(self.in_dist_train_dataloader.dataset.classes)
model = get_network(args=args)
checkpoints_folder = os.path.join(self.checkpoint_path, most_recent_folder(self.checkpoint_path, fmt=DATE_FORMAT))
weights_file = best_acc_weights(checkpoints_folder)
model.load_state_dict(torch.load(os.path.join(checkpoints_folder, weights_file)))
return model
class VitClassifier(Classifier):
def __init__(self, output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names):
super().__init__(output_dir, classifier_cfg, logger, current_run_name, data_source, ood_class_names)
def get_vit_transforms(self):
processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k")
image_mean = processor.image_mean
image_std = processor.image_std
size = processor.size["height"]
# size=32
normalize = Normalize(mean=image_mean, std=image_std)
self.train_transforms = Compose(
[
RandomResizedCrop(size),
RandomHorizontalFlip(),
ToTensor(),
normalize,
]
)
self.val_transforms = Compose(
[
Resize(size),
CenterCrop(size),
ToTensor(),
normalize,
]
)
def initiate_trainer(self):
self.logger.info(f"Initializing classifier\n")
self.preperae_id_ood_datasets(self.data_source, self.ood_class_names)
self.get_vit_transforms()
self.in_dist_train_dataloader, self.in_dist_val_dataloader, self.in_dist_test_dataloader = create_dataloaders(
train_transforms=self.train_transforms, val_transforms=self.val_transforms,
data_paths={"train": os.path.join(self.data_source, "train_ood_id_split"),
"val": os.path.join(self.data_source, "val_ood_id_split"),
"test": os.path.join(self.data_source, "test_ood_id_split"),
}, dataset_type=DatasetType.IN_DISTRIBUTION, val_batch_size=self.classifier_cfg.val_batch_size,
train_batch_size=self.classifier_cfg.train_batch_size,
dataloader_num_workers=self.classifier_cfg.dataloader_num_workers,
use_weighted_sampler=self.classifier_cfg.weighted_sampler)
self.id2label = self.in_dist_train_dataloader.dataset.labels_to_classes_names
self.label2id = self.in_dist_train_dataloader.dataset.classes_names_to_labels
self.model = ViTLightningModule(train_dataloader=self.in_dist_train_dataloader,
val_dataloader=self.in_dist_val_dataloader,
test_dataloader=self.in_dist_test_dataloader,
id2label=self.id2label,
label2id=self.label2id, num_labels=len(self.id2label),
model_path=self.classifier_cfg.model_path,
loss_class_weights=self.classifier_cfg.loss_class_weights, logger=self.logger,
max_epochs=self.classifier_cfg.max_epoch)
self.model = self.model.to(device)
def train(self):
self.initiate_trainer()
if self.classifier_cfg.retrain:
self.logger.info(f"Starting to train the ViT classifier\n")
early_stop_callback = EarlyStopping(
monitor='validation_loss',
patience=3,
strict=True,
verbose=True,
mode='min'
)
checkpoint_callback = ModelCheckpoint(
monitor='validation_loss',
mode='min',
save_top_k=1,
dirpath=self.checkpoint_path,
filename='best_model',
)
logger = TensorBoardLogger(save_dir=os.path.join(self.train_output_dir, "tb_logs"), name=self.current_run_name)
trainer = Trainer(num_nodes=1, max_epochs=self.classifier_cfg.max_epoch,
callbacks=[early_stop_callback, checkpoint_callback], logger=logger)
ckpt_path = os.path.join(self.checkpoint_path, "best_model.ckpt") if self.classifier_cfg.resume else None
trainer.fit(self.model, ckpt_path=ckpt_path)
"""Finally, let's test the trained model on the test set:"""
results = trainer.test()
print(results)
def get_fine_tuned_model(self):
model = ViTLightningModule.load_from_checkpoint(
os.path.join(self.checkpoint_path, f"best_model.ckpt"),
train_dataloader=self.in_dist_train_dataloader,
val_dataloader=self.in_dist_val_dataloader,
test_dataloader=self.in_dist_test_dataloader,
id2label=self.id2label,
label2id=self.label2id,
num_labels=len(self.label2id)
)
return model