-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
339 lines (254 loc) · 10.4 KB
/
train.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
import os
import torch
import numpy as np
from torchvision import datasets, transforms, models
import torch.optim as optim
import facenet_pytorch
from facenet_pytorch import MTCNN
from tqdm import tqdm
import torch.nn as nn
import random
import argparse
from collections import OrderedDict
from loadOpenFace import prepareOpenFace
### PARAMETERS
data_dir = 'dataset/cropped_images'
n_epochs = 10
chk_path = 'models/AvengersClassifier.pth' # Default checkpoint path
###
### Parse Arguments
parser = argparse.ArgumentParser(
description='Trains the model and saves the model')
parser.add_argument('-p', '--path', default=chk_path, help='Checkpoint path')
parser.add_argument('-d', '--dataset', default=data_dir, help='Dataset path')
parser.add_argument('-e', '--epochs', type=int, default=n_epochs, help='Number of Epochs')
args = parser.parse_args()
chk_path = args.path
n_epochs = args.epochs
data_dir = args.dataset
### Check if CUDA GPU is available
useCuda = torch.cuda.is_available()
if useCuda:
print('CUDA is avialable')
device = torch.device('cuda:0')
else:
print('CUDA is not avialable')
device = torch.device('cpu')
### Use MTCNN to Crop and Align Images
import warnings
warnings.filterwarnings("ignore", category=np.VisibleDeprecationWarning)
mtcnn = MTCNN(
image_size=160, margin=0, min_face_size=20,
thresholds=[0.6, 0.7, 0.7], factor=0.709, post_process=True,
device=device
)
aligned_data_dir = data_dir + '_aligned'
dataset = datasets.ImageFolder(data_dir, transform=transforms.Resize((512, 512)))
dataset.idx_to_class = {i:c for c, i in dataset.class_to_idx.items()}
# Replace the class label with the new path for storing aligned data
dataset.samples = [(p, p.replace(data_dir, aligned_data_dir)) for p, _ in dataset.samples]
batch_size = 32
num_workers = 0 if os.name == 'nt' else 8
dataloader = torch.utils.data.DataLoader(dataset,
batch_size=batch_size,
num_workers=num_workers,
collate_fn=facenet_pytorch.training.collate_pil)
# Run MTCNN for all the images and save them in new directory
for i, (image, path) in enumerate(tqdm(dataloader, desc="Converting")):
mtcnn(image, save_path=path)
# Delete to save memory
del mtcnn
del dataloader
print()
#### Augmenting the Dataset
class AugmentDataset(datasets.ImageFolder):
def __init__(self, root, transform = None):
super().__init__(root, transform)
self.all_labels = [int(x[1]) for x in self.imgs]
self.horizontalTransform = transforms.RandomHorizontalFlip(1)
def __len__(self):
return 2 * super().__len__()
def __getitem__(self, item):
if item < super().__len__():
image, label = super().__getitem__(item)
else:
item -= super().__len__()
image, label = super().__getitem__(item)
image = self.horizontalTransform(image)
return image, label
transform = transforms.Compose([transforms.Resize(96),
transforms.ToTensor()])
dataset = AugmentDataset(aligned_data_dir, transform=transform)
idx_to_class = {i:c for c, i in dataset.class_to_idx.items()}
total_count = len(dataset)
train_count = int(0.8 * total_count)
test_count = total_count - train_count
train_dataset, test_dataset = torch.utils.data.random_split(dataset,
[train_count, test_count])
print('Total Images : ', total_count)
print('Num of Train Images : ', len(train_dataset))
print('Num of Test Images : ', len(test_dataset))
print()
batch_size = 64
num_workers = 0 if os.name == 'nt' else 8
train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size,
num_workers=num_workers, shuffle=True)
test_dataloader = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size,
num_workers=num_workers, shuffle=True)
### Generate triplets Function
def generate_triplets(images, labels):
positive_images = []
negative_images = []
batch_size = len(labels)
for i in range(batch_size):
anchor_label = labels[i]
positive_list = []
negative_list = []
for j in range(batch_size):
if j != i:
if labels[j] == anchor_label:
positive_list.append(j)
else:
negative_list.append(j)
positive_images.append(images[random.choice(positive_list)])
negative_images.append(images[random.choice(negative_list)])
positive_images = torch.stack(positive_images)
negative_images = torch.stack(negative_images)
return positive_images, negative_images
### Define Triplet Loss
class TripletLoss(nn.Module):
def __init__(self, alpha=0.2):
super(TripletLoss, self).__init__()
self.alpha = alpha
def calc_euclidean(self, x1, x2):
return (x1 - x2).pow(2).sum(1)
def forward(self, anchor, positive, negative): # (batch_size , emb_size)
distance_positive = self.calc_euclidean(anchor, positive)
distance_negative = self.calc_euclidean(anchor, negative)
losses = torch.relu(distance_positive - distance_negative + self.alpha)
return losses.mean()
# Load inception model
model = prepareOpenFace(useCuda)
model.eval()
print("Inception Model Loaded")
# Define optimizer and loss for inception model
optimizer = optim.Adam(model.parameters(), lr=0.001)
loss_fn = TripletLoss()
# Training the inception model
for epoch in range(n_epochs):
train_loss = 0
count = 0
## Training Loop
model.train()
for batch, (images, labels) in enumerate(tqdm(train_dataloader, \
desc="Training", leave=False)):
positives , negatives = generate_triplets(images, labels)
# Move tensor to device
images, labels = images.to(device), labels.to(device)
positives, negatives = positives.to(device), negatives.to(device)
optimizer.zero_grad()
# Seaseme Network
anchor_out = model(images)
positive_out = model(positives)
negative_out = model(negatives)
# Get the loss
loss = loss_fn(anchor_out, positive_out, negative_out)
loss.backward()
optimizer.step()
train_loss += loss.detach().item()
count = len(labels)
print('Epoch : %d/%d - Loss: %0.4f' %
(epoch+1, n_epochs, train_loss / count))
train_loss = 0.0
model.eval()
print("Inception Model : Training Done\n")
### Transfer Learning the classifier
n_classes = len(dataset.class_to_idx)
# Define the classifier model
classifier_model = nn.Sequential(OrderedDict([
("nn4_small_v2", model),
("fc", nn.Linear(736, n_classes))
]))
classifier_model = classifier_model.to(device)
# Freeze the parameters in the nn4_small_v2 layer
for param in classifier_model.parameters():
param.requires_grad = False
for param in classifier_model.fc.parameters():
param.requires_grad = True
# Define optimizer and loss for classifier model
optimizer = optim.Adam(classifier_model.fc.parameters(), lr=0.01)
loss_fn = torch.nn.CrossEntropyLoss()
### Training the Classifier
print("Training Classifier")
def train(n_epochs, dataloader, model, optimizer, loss_fn):
'''returns Trained classifier model'''
for epoch in range(n_epochs):
train_loss = 0.0
count = 0
# Training loop
model.train()
for batch, (images, labels) in enumerate(tqdm(dataloader, \
desc="Training", leave=False)):
# Move Tensor to appropriate device
images, labels = images.to(device), labels.to(device)
optimizer.zero_grad()
out = model(images)
# Get the loss
loss = loss_fn(out, labels)
loss.backward()
optimizer.step()
train_loss += loss.detach().item()
count = len(labels)
print('Epoch : %d/%d - Loss: %0.4f' %
(epoch+1, n_epochs, train_loss / count))
train_loss = 0.0
model.eval()
print("Classifier Model : Training Done\n")
return model
# call the train function
classifier_model = train(10 , train_dataloader, classifier_model, optimizer, loss_fn)
### Testing the classifier
def test(dataloader, model, loss_fn):
test_loss = 0.0
total = 0
correct = 0
# Testing loop
model.eval()
for batch, (images, labels) in enumerate(tqdm(dataloader, \
desc="Testing")):
# Move Tensor to appropriate device
images, labels = images.to(device), labels.to(device)
with torch.no_grad():
out = model(images)
loss = loss_fn(out, labels)
test_loss += loss.detach().item()
# Get the class with max probability
pred = out.data.max(1, keepdim=True)[1]
# Compare predictions with true label
correct += np.sum(np.squeeze(pred.eq(labels.view_as(pred))).cpu().numpy())
total += labels.size(0)
print('Test Loss: {:.6f}\n'.format(test_loss/total))
print('Test Accuracy : %d%% (%d/%d)' % (
100 * correct / total, correct, total))
print()
return(float(correct / total))
# call the test function
current_accuracy = test(test_dataloader, classifier_model, loss_fn)
### Define Function to save model
def save_model(model, chk_path, idx_to_class, current_accuracy=1.0):
'''Saves the model only if model doesnt exist or
if the previous model accuracy was better'''
try:
checkpoint = torch.load(chk_path, map_location=torch.device('cpu'))
if(current_accuracy < checkpoint['accuracy']):
print("Not Saving, Previous model was better")
return
except FileNotFoundError:
print("Previous model not found")
torch.save({
'model_state_dict' : model.state_dict(),
'accuracy' : current_accuracy,
'idx_to_class': idx_to_class
}, chk_path)
print("Model Saved : %s" % chk_path)
save_model(classifier_model, chk_path, idx_to_class, current_accuracy)