-
Notifications
You must be signed in to change notification settings - Fork 9
/
fastai_cnn_learner.py
134 lines (106 loc) · 4.44 KB
/
fastai_cnn_learner.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
# Import linraries
import os
import urllib
import shutil
from tqdm import tqdm
# from sklearn.metrics import confusion_matrix
from fastai import *
from fastai.vision import *
from fastai.callbacks import *
from optimizers.optimizers import Ranger
import torch
import warnings
warnings.filterwarnings('ignore')
def train_fit_fc():
# train head
learn.fit_fc(10,
lr,
start_pct=0.7,
callbacks = [SaveModelCallback(learn, every='improvement', monitor='accuracy',
name='stage1')])
# load stage1
learn.load("stage1")
learn.unfreeze()
learn.fit_fc(40,
lr/10, # try between [lr/10, lr/50]
start_pct=0.7,
callbacks = [SaveModelCallback(learn, every='improvement', monitor='accuracy',
name='stage2'),
EarlyStoppingCallback(learn, monitor="accuracy", patience = 10)])
# further training
learn.load("stage2")
learn.export("/home/max/project/stage2-checkpt")
learn.unfreeze()
learn.fit_fc(60,
lr/100, # try between [lr/50, lr/100]
start_pct=0.7,
callbacks = [SaveModelCallback(learn, every='improvement', monitor='accuracy',
name='stage3'),
EarlyStoppingCallback(learn, monitor="accuracy", patience = 15)])
learn.load("stage3")
learn.export("/home/max/project/stage3-checkpt")
def train_one_cycle():
# train head
learn.fit_one_cycle(10,
lr,
callbacks = [SaveModelCallback(learn, every='improvement', monitor='accuracy',
name='stage1-onecycle')])
# load stage1
learn.load("stage1-onecycle")
# training full model
learn.unfreeze()
learn.fit_one_cycle(10,
max_lr=slice(lr/100,lr/2), # try between [lr/2, lr/10]
callbacks = [SaveModelCallback(learn, every='improvement', monitor='accuracy',
name='stage2-onecycle')])
# stage3 further training
learn.load("stage2-onecycle")
learn.export("/home/max/project/stage2-onecycle-checkpt")
learn.unfreeze()
learn.fit_one_cycle(15,
max_lr=slice(lr/100,lr/100), # try between [lr/10, lr/100]
callbacks = [SaveModelCallback(learn, every='improvement', monitor='accuracy',
name='stage3-onecycle')])
# exporting models
learn.load("stage3-onecycle")
learn.export("/home/max/project/stage3-onecycle-checkpt")
if __name__ == "__main__":
torch.cuda.set_device(0)
# define image data directory path
DATA_DIR='./occasion_data'
# The directory under the path is the label name.
os.listdir(f'{DATA_DIR}')
torch.cuda.is_available()
tfms = get_transforms(max_rotate=20, max_zoom=1.3, max_lighting=0.4, max_warp=0.4,
p_affine=1., p_lighting=1.)
# create image data bunch
data = ImageDataBunch.from_folder(DATA_DIR,
train=".",
valid_pct=0.2,
ds_tfms=tfms,
size=224,
padding_mode = "reflection",
bs=64,
num_workers=0).normalize(imagenet_stats)
# check classes
print(f'Classes: \n {data.classes}')
# build model (use resnet34)
learn = cnn_learner(data,
models.resnet50,
metrics=accuracy,
model_dir="/home/max/gender_classification/model/",
loss_func=LabelSmoothingCrossEntropy(),
opt_func = partial(Ranger),
true_wd = True,
bn_wd = False)
learn.mixup()
# find lr
# can comment this out in later runs
learn.lr_find()
lr_find = learn.recorder.plot(suggestion = True, skip_end=15, return_fig = True)
lr_find.savefig("lr_find.png")
lr = 3e-2
learn.to_fp16()
# gen_data()
train_fit_fc()
# train_one_cycle()