-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodel.py
230 lines (187 loc) · 9.66 KB
/
model.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
import torch
from torch import nn
import pdb
from models import resnet, resnext, resnetl, c3d, mobilenetv2, shufflenetv2
def generate_model(opt):
assert opt.model in [
'resnet', 'resnetl', 'resnext', 'c3d', 'mobilenetv2', 'shufflenetv2'
]
if opt.model == 'resnet':
assert opt.model_depth in [10, 50]
from models.resnet import get_fine_tuning_parameters
if opt.model_depth == 10:
model = resnet.resnet10(
num_classes=opt.n_classes,
shortcut_type=opt.resnet_shortcut,
sample_size=opt.sample_size,
sample_duration=opt.sample_duration)
elif opt.model_depth == 50:
model = resnet.resnet50(
num_classes=opt.n_classes,
shortcut_type=opt.resnet_shortcut,
sample_size=opt.sample_size,
sample_duration=opt.sample_duration)
elif opt.model == 'resnetl':
assert opt.model_depth in [10, 18]
from models.resnetl import get_fine_tuning_parameters
if opt.model_depth == 10:
model = resnetl.resnetl10(
num_classes=opt.n_classes,
shortcut_type=opt.resnet_shortcut,
sample_size=opt.sample_size,
sample_duration=opt.sample_duration)
elif opt.model_depth == 18:
model = resnetl.resnetl10(
num_classes=opt.n_classes,
shortcut_type=opt.resnet_shortcut,
sample_size=opt.sample_size,
sample_duration=opt.sample_duration)
elif opt.model == 'resnext':
assert opt.model_depth in [101]
from models.resnext import get_fine_tuning_parameters
if opt.model_depth == 101:
model = resnext.resnet101(
num_classes=opt.n_classes,
shortcut_type=opt.resnet_shortcut,
cardinality=opt.resnext_cardinality,
sample_size=opt.sample_size,
sample_duration=opt.sample_duration)
elif opt.model == 'c3d':
assert opt.model_depth in [10]
from models.c3d import get_fine_tuning_parameters
if opt.model_depth == 10:
model = c3d.c3d_v1(
sample_size=opt.sample_size,
sample_duration=opt.sample_duration,
num_classes=opt.n_classes)
elif opt.model == 'mobilenetv2':
from models.mobilenetv2 import get_fine_tuning_parameters
model = mobilenetv2.mob_v2(
num_classes=opt.n_classes,
sample_size=opt.sample_size,
width_mult=opt.width_mult)
elif opt.model == 'shufflenetv2':
from models.shufflenetv2 import get_fine_tuning_parameters
model = shufflenetv2.shf_v2(
num_classes=opt.n_classes,
sample_size=opt.sample_size,
width_mult=opt.width_mult)
if not opt.no_cuda:
model = nn.DataParallel(model, device_ids=None)
if opt.pretrain_path:
print('loading pretrained model {}'.format(opt.pretrain_path))
pretrain = torch.load(opt.pretrain_path)
assert opt.arch == pretrain['arch']
if opt.pretrain_dataset == 'jester':
if opt.sample_duration < 32 and opt.model != 'c3d':
model = _modify_first_conv_layer(model,3,3)
if opt.model in ['mobilenetv2', 'shufflenetv2']:
del pretrain['state_dict']['module.classifier.1.weight']
del pretrain['state_dict']['module.classifier.1.bias']
else:
del pretrain['state_dict']['module.fc.weight']
del pretrain['state_dict']['module.fc.bias']
model.load_state_dict(pretrain['state_dict'],strict=False)
if opt.modality in ['RGB', 'flo'] and opt.model != 'c3d':
print("[INFO]: RGB model is used for init model")
if opt.dataset != 'jester' and not opt.no_first_lay:
model = _modify_first_conv_layer(model,3,3) ##### Check models trained (3,7,7) or (7,7,7)
elif opt.modality in ['Depth', 'seg']:
print("[INFO]: Converting the pretrained model to Depth init model")
model = _construct_depth_model(model)
print("[INFO]: Done. Flow model ready.")
elif opt.modality in ['RGB-D', 'RGB-flo', 'RGB-seg']:
print("[INFO]: Converting the pretrained model to RGB+D init model")
model = _construct_rgbdepth_model(model)
if opt.no_first_lay:
model = _modify_first_conv_layer(model,3,4) ##### Check models trained (3,7,7) or (7,7,7)
print("[INFO]: Done. RGB-D model ready.")
if opt.pretrain_dataset == opt.dataset:
model.load_state_dict(pretrain['state_dict'])
elif opt.pretrain_dataset in ['egogesture', 'nv', 'denso']:
del pretrain['state_dict']['module.fc.weight']
del pretrain['state_dict']['module.fc.bias']
model.load_state_dict(pretrain['state_dict'],strict=False)
# Check first kernel size
modules = list(model.modules())
first_conv_idx = list(filter(lambda x: isinstance(modules[x], nn.Conv3d),
list(range(len(modules)))))[0]
conv_layer = modules[first_conv_idx]
if conv_layer.kernel_size[0]> opt.sample_duration:
print("[INFO]: RGB model is used for init model")
model = _modify_first_conv_layer(model,int(opt.sample_duration/2),1)
if opt.model == 'c3d':# CHECK HERE
model.module.fc = nn.Linear(
model.module.fc[0].in_features, model.module.fc[0].out_features)
model.module.fc = model.module.fc.cuda()
elif opt.model in ['mobilenetv2', 'shufflenetv2']:
model.module.classifier = nn.Sequential(
nn.Dropout(0.9),
nn.Linear(model.module.classifier[1].in_features, opt.n_finetune_classes))
model.module.classifier = model.module.classifier.cuda()
else:
model.module.fc = nn.Linear(model.module.fc.in_features,
opt.n_finetune_classes)
model.module.fc = model.module.fc.cuda()
parameters = get_fine_tuning_parameters(model, opt.ft_begin_index)
model = model.cuda()
return model, parameters
else:
print('ERROR no cuda')
return model, model.parameters()
def _construct_depth_model(base_model):
# modify the first convolution kernels for Depth input
modules = list(base_model.modules())
first_conv_idx = list(filter(lambda x: isinstance(modules[x], nn.Conv3d),
list(range(len(modules)))))[0]
conv_layer = modules[first_conv_idx]
container = modules[first_conv_idx - 1]
# modify parameters, assume the first blob contains the convolution kernels
motion_length = 1
params = [x.clone() for x in conv_layer.parameters()]
kernel_size = params[0].size()
new_kernel_size = kernel_size[:1] + (1*motion_length, ) + kernel_size[2:]
new_kernels = params[0].data.mean(dim=1, keepdim=True).expand(new_kernel_size).contiguous()
new_conv = nn.Conv3d(1, conv_layer.out_channels, conv_layer.kernel_size, conv_layer.stride,
conv_layer.padding, bias=True if len(params) == 2 else False)
new_conv.weight.data = new_kernels
if len(params) == 2:
new_conv.bias.data = params[1].data # add bias if neccessary
layer_name = list(container.state_dict().keys())[0][:-7] # remove .weight suffix to get the layer name
# replace the first convlution layer
setattr(container, layer_name, new_conv)
return base_model
def _construct_rgbdepth_model(base_model):
# modify the first convolution kernels for RGB-D input
modules = list(base_model.modules())
first_conv_idx = list(filter(lambda x: isinstance(modules[x], nn.Conv3d),
list(range(len(modules)))))[0]
conv_layer = modules[first_conv_idx]
container = modules[first_conv_idx - 1]
# modify parameters, assume the first blob contains the convolution kernels
motion_length = 1
params = [x.clone() for x in conv_layer.parameters()]
kernel_size = params[0].size()
new_kernel_size = kernel_size[:1] + (1 * motion_length,) + kernel_size[2:]
new_kernels = torch.mul(torch.cat((params[0].data, params[0].data.mean(dim=1,keepdim=True).expand(new_kernel_size).contiguous()), 1), 0.6)
new_kernel_size = kernel_size[:1] + (3 + 1 * motion_length,) + kernel_size[2:]
new_conv = nn.Conv3d(4, conv_layer.out_channels, conv_layer.kernel_size, conv_layer.stride,
conv_layer.padding, bias=True if len(params) == 2 else False)
new_conv.weight.data = new_kernels
if len(params) == 2:
new_conv.bias.data = params[1].data # add bias if neccessary
layer_name = list(container.state_dict().keys())[0][:-7] # remove .weight suffix to get the layer name
# replace the first convolution layer
setattr(container, layer_name, new_conv)
return base_model
def _modify_first_conv_layer(base_model, new_kernel_size1, new_filter_num):
modules = list(base_model.modules())
first_conv_idx = list(filter(lambda x: isinstance(modules[x], nn.Conv3d),
list(range(len(modules)))))[0]
conv_layer = modules[first_conv_idx]
container = modules[first_conv_idx - 1]
new_conv = nn.Conv3d(new_filter_num, conv_layer.out_channels, kernel_size=(new_kernel_size1,7,7),
stride=(1,2,2), padding=(1,3,3), bias=False)
layer_name = list(container.state_dict().keys())[0][:-7]
setattr(container, layer_name, new_conv)
return base_model