-
Notifications
You must be signed in to change notification settings - Fork 1
/
save_plk.py
214 lines (185 loc) · 8.46 KB
/
save_plk.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
from __future__ import print_function
from PIL import Image, ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
import argparse
import csv
import os
import collections
import pickle
import random
import numpy as np
import torch
from torch.autograd import Variable
import torch.backends.cudnn as cudnn
import torch.nn as nn
import torch.optim as optim
import torchvision.transforms as transforms
import torchvision.datasets as datasets
from io_utils import model_dict, parse_args
from data.datamgr import SimpleDataManager , SetDataManager
import configs
import copy
from methods.baselinetrain import BaselineTrain
import wrn_mixup_model # wrn_model
import torch.nn.functional as F
from io_utils import parse_args, get_resume_file ,get_assigned_file
from os import path
use_gpu = torch.cuda.is_available()
class WrappedModel(nn.Module):
def __init__(self, module):
super(WrappedModel, self).__init__()
self.module = module
def forward(self, x):
return self.module(x)
def save_pickle(file, data):
with open(file, 'wb') as f:
pickle.dump(data, f)
def load_pickle(file):
with open(file, 'rb') as f:
return pickle.load(f)
def extract_feature(val_loader, model, checkpoint_dir,
dataname, dgr="0",
tag='last',set='base',
redo=False,
flip=False,
img_size=84):
'''
dgr = "0" # "0" | 90 | 180 | 270
redo: generate feature again
'''
save_dir = '{}/{}'.format(checkpoint_dir, tag)
# feat_file_name = '{}/{}_features.plk{}'.format(save_dir, set, dgr)
feat_file_name = '{}/{}_features_s{}_r{}_f{}.plk'.format(save_dir,
set,
img_size,
'0' if not dgr else dgr,
'1' if flip else '0')
print('>>> writing to {} for {} ...'.format(feat_file_name, dataname))
if os.path.isfile(feat_file_name):
if redo:
print('>>> found old features, re-generating...')
else:
data = load_pickle(feat_file_name)
return data
else:
if not os.path.isdir(save_dir):
os.makedirs(save_dir)
#model.eval()
with torch.no_grad():
output_dict = collections.defaultdict(list)
for i, (inputs, labels) in enumerate(val_loader):
print(i, '-', end='', flush=True)
# ~~~~~ rotation ~~~~~
# TODO: fix a bug about dgr '0'
if dgr == "0" and flip:
inputs = inputs.flip(2) # 0 degree, flip it
if dgr == "90":
if flip:
inputs = inputs.permute(0,1,3,2) # 90 degree, flip it
else:
inputs = inputs.permute(0,1,3,2).flip(2) # 90 degree
elif dgr == "180":
inputs = inputs.permute(0,1,3,2).flip(2) # 90 degree
if flip:
inputs = inputs.permute(0,1,3,2) # 180 degree, flip it
else:
inputs = inputs.permute(0,1,3,2).flip(2) # 180 degree
elif dgr == "270":
inputs = inputs.permute(0,1,3,2).flip(2) # 90 degree
inputs = inputs.permute(0,1,3,2).flip(2) # 180 degree
if flip:
inputs = inputs.permute(0,1,3,2) # 270 degree, flip it
else:
inputs = inputs.permute(0,1,3,2).flip(2) # 270 degree
if dgr not in ["0", "90", "180", "270"]:
print('ERROR degree!')
exit()
# ~~~~~ rotation ~~~~~
# compute output
if use_gpu: # inputs shape: torch.Size([256, 3, 84, 84])
inputs = inputs.cuda()
labels = labels.cuda()
outputs,_ = model(inputs)
outputs = outputs.cpu().data.numpy()
for out, label in zip(outputs, labels):
output_dict[label.item()].append(out)
all_info = output_dict
save_pickle(feat_file_name, all_info) # change feature file name here!
return all_info
if __name__ == '__main__':
params = parse_args('test')
## params.model = 'WideResNet28_10' # 'Conv4S' 'WideResNet28_10'
## params.method = 'S2M2_R' # 'S2M2_R' 'rotation' 'manifold_mixup'
loadfile_base = configs.data_dir[params.dataset] + 'base.json'
loadfile_novel = configs.data_dir[params.dataset] + 'novel.json'
loadfile_val = configs.data_dir[params.dataset] + 'val.json'
img_size = params.re_size
TF = {'True':True, 'False':False}
flip = TF[params.flip]
if params.dataset == 'miniImagenet' or params.dataset == 'CUB' or params.dataset == 'tieredImageNet':
datamgr = SimpleDataManager(img_size, batch_size = 100) # 80 is s2m2's and 84 is DC's default
elif params.dataset == 'cifar':
datamgr = SimpleDataManager(img_size, batch_size = 256) # 32 is default
elif params.dataset == 'MultiDigitMNIST':
datamgr = SimpleDataManager(img_size, batch_size = 256) # 64 is default
base_loader = datamgr.get_data_loader(loadfile_base, aug=False, shf=False)
novel_loader = datamgr.get_data_loader(loadfile_novel, aug = False, shf=False)
val_loader = datamgr.get_data_loader(loadfile_val, aug = False, shf=False)
checkpoint_dir = '%s/checkpoints/%s/%s_%s' %(configs.save_dir,
params.dataset,
params.model,
params.method)
print('checkpoint_dir', checkpoint_dir)
modelfile = get_resume_file(checkpoint_dir, fetch_epoch = params.fetch_epoch)
if params.model == 'WideResNet28_10':
if params.dataset == 'cifar':
model = wrn_mixup_model.wrn28_10(num_classes=64,
loss_type='softmax' if (params.method=='S2M2_R') else 'dist')
else:
model = wrn_mixup_model.wrn28_10(num_classes=params.num_classes)
elif params.model == 'Conv4S':
model = BaselineTrain( model_dict[params.model], params.num_classes, loss_type = 'softmax') # softmax | dist
if use_gpu:
model = model.cuda()
cudnn.benchmark = True
print('> loading model from:', modelfile)
checkpoint = torch.load(modelfile)
state = checkpoint['state']
state_keys = list(state.keys())
print('----------\n', state_keys, '\n----------\n')
callwrap = False
if 'module' in state_keys[0]:
callwrap = True
if callwrap:
model = WrappedModel(model)
this_model_dict = model.state_dict()
this_model_dict.update(state)
model_dict_ = copy.deepcopy(this_model_dict)
for key in this_model_dict:
if 'linear.' in key:
model_dict_[key.replace('linear.','classifier.')] = model_dict_.pop(key)
if params.method in ['rotation'] and params.dataset != 'MultiDigitMNIST':
for key in this_model_dict:
if 'feature.' in key:
model_dict_[key.replace('feature.','')] = model_dict_.pop(key)
print(list(model_dict_.keys()))
model.load_state_dict(model_dict_) # not this_model_dict
model.eval()
if 'b' in params.bvn:
output_dict_base = extract_feature(base_loader, model, checkpoint_dir,
params.dataset, dgr=params.dgr,
tag='last', set='base', redo=True, flip=flip,
img_size=img_size)
print("base set features saved!")
if 'v' in params.bvn:
output_dict_val = extract_feature(val_loader, model, checkpoint_dir,
params.dataset, dgr=params.dgr,
tag='last',set='val', redo=True, flip=flip,
img_size=img_size)
print("val features saved!")
if 'n' in params.bvn:
output_dict_novel = extract_feature(novel_loader, model, checkpoint_dir,
params.dataset, dgr=params.dgr,
tag='last',set='novel', redo=True, flip=flip,
img_size=img_size)
print("novel features saved!")