-
Notifications
You must be signed in to change notification settings - Fork 131
/
Copy pathresnet18_from_resnet34.yaml
153 lines (150 loc) · 4.28 KB
/
resnet18_from_resnet34.yaml
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
datasets:
&imagenet_train ilsvrc2012/train: !import_call
_name: &dataset_name 'ilsvrc2012'
_root: &root_dir !join ['~/datasets/', *dataset_name]
key: 'torchvision.datasets.ImageFolder'
init:
kwargs:
root: !join [*root_dir, '/train']
transform: !import_call
key: 'torchvision.transforms.Compose'
init:
kwargs:
transforms:
- !import_call
key: 'torchvision.transforms.RandomResizedCrop'
init:
kwargs:
size: &input_size [224, 224]
- !import_call
key: 'torchvision.transforms.RandomHorizontalFlip'
init:
kwargs:
p: 0.5
- !import_call
key: 'torchvision.transforms.ToTensor'
init:
- !import_call
key: 'torchvision.transforms.Normalize'
init:
kwargs: &normalize_kwargs
mean: [0.485, 0.456, 0.406]
std: [0.229, 0.224, 0.225]
&imagenet_val ilsvrc2012/val: !import_call
key: 'torchvision.datasets.ImageFolder'
init:
kwargs:
root: !join [*root_dir, '/val']
transform: !import_call
key: 'torchvision.transforms.Compose'
init:
kwargs:
transforms:
- !import_call
key: 'torchvision.transforms.Resize'
init:
kwargs:
size: 256
- !import_call
key: 'torchvision.transforms.CenterCrop'
init:
kwargs:
size: *input_size
- !import_call
key: 'torchvision.transforms.ToTensor'
init:
- !import_call
key: 'torchvision.transforms.Normalize'
init:
kwargs: *normalize_kwargs
models:
teacher_model:
key: &teacher_model_key 'resnet34'
_weights: &teacher_weights !import_get
key: 'torchvision.models.resnet.ResNet34_Weights'
kwargs:
num_classes: 1000
weights: !getattr [*teacher_weights, 'IMAGENET1K_V1']
src_ckpt:
student_model:
key: &student_model_key 'resnet18'
kwargs:
num_classes: 1000
_experiment: &student_experiment !join [*dataset_name, '-', *student_model_key, '_from_', *teacher_model_key]
src_ckpt:
dst_ckpt: !join ['./resource/ckpt/ilsvrc2012/kd_w_ls/', *student_experiment, '.pt']
train:
log_freq: 1000
num_epochs: 100
train_data_loader:
dataset_id: *imagenet_train
sampler:
class_or_func: !import_get
key: 'torch.utils.data.RandomSampler'
kwargs:
kwargs:
batch_size: 512
num_workers: 16
pin_memory: True
drop_last: False
cache_output:
val_data_loader:
dataset_id: *imagenet_val
sampler: &val_sampler
class_or_func: !import_get
key: 'torch.utils.data.SequentialSampler'
kwargs:
kwargs:
batch_size: 32
num_workers: 16
pin_memory: True
drop_last: False
teacher:
forward_proc: 'forward_batch_only'
sequential: []
wrapper: 'DataParallel'
requires_grad: False
student:
forward_proc: 'forward_batch_only'
adaptations:
sequential: []
wrapper: 'DistributedDataParallel'
requires_grad: True
frozen_modules: []
optimizer:
key: 'SGD'
kwargs:
lr: 0.2
momentum: 0.9
weight_decay: 0.0001
scheduler:
key: 'MultiStepLR'
kwargs:
milestones: [30, 60, 90]
gamma: 0.1
criterion:
key: 'WeightedSumLoss'
kwargs:
sub_terms:
kd:
criterion:
key: 'LogitStdKDLoss'
kwargs:
student_module_path: '.'
student_module_io: 'output'
teacher_module_path: '.'
teacher_module_io: 'output'
temperature: 2.0
alpha: 0.5
beta: 9
reduction: 'batchmean'
weight: 1.0
test:
test_data_loader:
dataset_id: *imagenet_val
sampler: *val_sampler
kwargs:
batch_size: 1
num_workers: 16
pin_memory: True
drop_last: False