-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
162 lines (133 loc) · 6.32 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
import pandas as pd
import numpy as np
import torch
import os
from PIL import Image
import torchvision
from torch.utils.tensorboard import SummaryWriter
import matplotlib.pyplot as plt
from tqdm import tqdm
from CityscapesDataloader import CityscapesDataset
from UNet import UNet
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
device = torch.device('mps' if torch.backends.mps.is_available() else 'cpu')
# create models folder if it doesn't exist
if not os.path.exists('checkpoints'):
os.makedirs('checkpoints')
# Hyperparameters
masked_aux = False
max_epochs = 100
num_classes = 8
batch_size = 8
img_size = (256, 256)
learning_rate = 0.0001
def main():
# Load Data
train_dataset1 = CityscapesDataset(split='train', img_size=img_size, rotate=True)
train_dataset2 = CityscapesDataset(split='train', img_size=img_size, rotate=True, foggy=True)
train_dataset = torch.utils.data.ConcatDataset([train_dataset1, train_dataset2])
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True, num_workers=4)
test_dataset = CityscapesDataset(split='test', img_size=img_size)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False)
# Model
model = UNet(num_class=num_classes, out_sz=img_size, masked_aux=masked_aux).to(device)
# print number of parameters
print(f'Number of parameters: {sum(p.numel() for p in model.parameters())}')
# model is pytorch resnet
# model = torchvision.models.resnet18(pretrained=False)
# model.fc = torch.nn.Linear(512, 4)
# model = model.to(device)
# load model if it exists
# if os.path.exists('checkpoints/model-7000.ckpt'):
# model.load_state_dict(torch.load('checkpoints/model-7000.ckpt'))
# print('Model loaded from checkpoints/model.ckpt')
# Loss and optimizer
criterion = torch.nn.CrossEntropyLoss()
criterion_mse = torch.nn.MSELoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
# Create tensorboard writer
writer = SummaryWriter()
# Train the model
log_steps = 50
class_loss_total = 0
rot_loss_total = 0
total_step = len(train_loader)
step = 0
for epoch in range(max_epochs):
for i, (images, labels, rot_label) in enumerate(train_loader):
images = images.to(device)
labels = labels.to(device)
rot_label = rot_label.to(device)
# import IPython ; IPython.embed() ; exit(1)
if masked_aux:
images_label = images
# mask = train_dataset.sample_mask().to(device)
# images = images * (1 - mask)
# Forward pass
outputs = model(images)
# rot_pred = outputs
out_pred = outputs[0]
rot_pred = outputs[1]
class_loss = criterion(out_pred, labels)
if masked_aux:
rot_loss = criterion_mse(rot_pred, images_label)
else:
rot_loss = criterion(rot_pred, rot_label)
# loss = class_loss + rot_loss
loss = class_loss
# Backward and optimize
optimizer.zero_grad()
loss.backward()
optimizer.step()
# Track the loss
class_loss_total += class_loss.item()
# class_loss_total = 0
rot_loss_total += rot_loss.item()
step += 1
if step % log_steps == 0:
total_loss = class_loss_total + rot_loss_total
writer.add_scalar('class_loss', class_loss_total / log_steps, epoch * total_step + i)
writer.add_scalar('rot_loss', rot_loss_total / log_steps, epoch * total_step + i)
writer.add_scalar('total_loss', total_loss / log_steps, epoch * total_step + i)
# print loss including rotation loss and class loss
print('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}, Class Loss: {:.4f}, Rot Loss: {:.4f}'
.format(epoch + 1, max_epochs, (step % total_step), total_step, total_loss / log_steps, class_loss_total / log_steps, rot_loss_total / log_steps))
# print(torch.softmax(rot_pred, dim=1))
# print(rot_label)
class_loss_total = 0
rot_loss_total = 0
if step % 2500 == 0:
# evaluate train accuracy
total_accuracy = 0
train_dataset.rotate = False
with torch.no_grad():
for i, (images, labels, rot_label) in enumerate(tqdm(train_loader)):
images = images.to(device)
labels = labels.to(device)
rot_label = rot_label.to(device)
outputs = model(images)
out_pred = outputs[0]
rot_pred = outputs[1]
class_labels = torch.Tensor(np.argmax(labels.cpu().numpy(), 1)).to(device)
_, predicted = torch.max(out_pred.data, 1)
total_accuracy += (predicted == class_labels).sum().item() / (img_size[0] * img_size[1])
train_accuracy = total_accuracy / len(train_dataset)
print('Train Accuracy: {:.4f}'.format(train_accuracy))
writer.add_scalar('train_accuracy', train_accuracy, step)
torch.save(model.state_dict(), os.path.join('checkpoints', 'model-{}.ckpt'.format(step)))
if train_accuracy > 0.9:
print("Final step: {}".format(step))
print("Final train accuracy: {}".format(train_accuracy))
exit()
train_dataset.rotate = True
print("Final step: {}".format(step))
print("Final train accuracy: {}".format(train_accuracy))
torch.save(model.state_dict(), os.path.join('checkpoints', 'model-{}.ckpt'.format(step)))
if __name__ == "__main__":
main()
# model-27500-segonly-rotaug.ckpt : 90.06% train accuracy - epoch 75
# model-32500-joint-auto : Train Accuracy: 0.9026 - epoch 88
# model-37200-joint-auto-mask : Train Accuracy: 0.8953 - epoch 100
# model-35000-segonly-clearfog.ckpt : 89.12% train accuracy - epoch 50 (100)
# model-32500-segonly-rotaug.ckpt : 90.26% train accuracy - epoch 88
# model-32500-joint-auto.ckpt : 90.06% train accuracy - epoch 88