-
Notifications
You must be signed in to change notification settings - Fork 4
/
PredNetModel.py
238 lines (212 loc) · 8.77 KB
/
PredNetModel.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
229
230
231
232
233
234
235
236
237
import torch
from torch import nn
import torch.nn.functional as f
from torch.autograd import Variable
from Encoder import Encoder
from Decoder import Decoder
import numpy as np
import os
import cv2
class PrednetModel(nn.Module):
def __init__(self,number_of_layers,\
image_size,numCh,\
SaveModelPath,\
OpenModelPath,\
numSaveIter):
super(PrednetModel, self).__init__()
self.image_size = image_size
self.input_layer_size = (3, 6, 32, 64, 128, 256, 512 )
self.hidden_size = (3, 16, 32, 64, 128, 256, 512 )
self.Elt_size = (6, 32, 64, 128, 256, 512 )
self.SaveModelPath = SaveModelPath
self.OpenModelPath = OpenModelPath
self.numSaveIter = numSaveIter
self.kernel_size = numCh
self.number_of_layers = number_of_layers
self.error_states = self.get_init_Elt_state()
self.saveModel = False
self.openModel = False
self.stateCheck()
def stateCheck(self):
if self.SaveModelPath != "":
print("\n ==> Save model at: " + self.SaveModelPath)
self.saveModel = True
if self.OpenModelPath != "":
print("\n ==> Open model at: " + self.SaveModelPath)
self.openModel = True
def hidden_layers_selctor(self,nlayer):
h_l_down_in = self.input_layer_size[nlayer]
h_l_top_out = self.hidden_size [nlayer]
h_l_down_out = self.hidden_size [nlayer]
h_Elt = self.Elt_size [nlayer]
return h_l_down_in,h_l_top_out,\
h_l_down_out,h_Elt
def get_init_Elt_state(self):
errot_list = []
pooling = self.image_size
for i in xrange(self.number_of_layers-1):
pooling = pooling/2
for layer in xrange(self.number_of_layers + 1):
errot_list.append([1,self.input_layer_size[layer],\
pooling*2**(self.number_of_layers-layer),\
pooling*2**(self.number_of_layers-layer)])
return errot_list
def get_init_Elt_tensor(self,state):
return Variable(torch.randn(state[0],state[1],state[2],state[3])).cuda()
def save_models(self,model,epoch,typeModel):
torch.save(model,\
os.path.join(self.SaveModelPath,\
typeModel+"_epoch_"+str(epoch)+'.pt'))
def call_Decoder(self,nlayer,Elt_state,\
Rlt_top,Rlt_state,epoch):
# h = hidden layers
R_lt_next = None
Elt_state_in = None
Decode_lt = None
# return hidden layers size:
h_l_down_in, h_l_top_out,\
h_l_down_out, h_Elt = self.hidden_layers_selctor(nlayer)
if Elt_state is None:
Elt_state_in = self.get_init_Elt_tensor(self.error_states[nlayer+1])
else:
Elt_state_in = Elt_state
if Rlt_top is None:
Decode_lt = Decoder(h_Elt, 0, h_l_down_out,\
self.kernel_size).cuda()
Elt_state = self.get_init_Elt_tensor(self.error_states[nlayer+1])
R_lt_next = Decode_lt(Elt_state_in,None,None)
else:
Decode_lt = Decoder(h_Elt, Rlt_top.data.size()[1],\
h_l_top_out,self.kernel_size).cuda()
R_lt_next = Decode_lt(Elt_state_in,\
Rlt_top,Rlt_state)
if self.saveModel == True:
if epoch%self.numSaveIter == 0:
self.save_models(Decode_lt,epoch,"Decoder")
return R_lt_next,Decode_lt.parameters()
def call_Encoder(self,nlayer,\
x_lt,R_lt,first,epoch):
E_lt = None
Encode_lt = None
# return hidden layers size:
h_l_down_in, h_l_top_out,\
h_l_down_out, h_Elt = self.hidden_layers_selctor(nlayer)
Encode_lt = Encoder([h_l_down_in,h_l_down_out],\
h_l_down_out,self.kernel_size,\
self.image_size).cuda()
if first is True:
E_lt = Encode_lt(x_lt,R_lt,True)
else:
E_lt = Encode_lt(x_lt,R_lt,False)
if self.saveModel == True:
if epoch%self.numSaveIter == 0:
self.save_models(Encode_lt,epoch,"Encoder")
return E_lt,Encode_lt.parameters()
## Algorithm 1 Calculation of PredNet states (page 4 paper)
def forward(self,x_t,Elt_state,\
Rlt_state,epoch):
Rlt = None
Rlt_top = None
Elt_prev = None
parD = None
parG = None
pGlist = nn.ParameterList()
pDlist = nn.ParameterList()
Rlt_state_tmp = [None] * (self.number_of_layers)
Elt_state_tmp = [None] * (self.number_of_layers)
# 1.a) Generative part:
for layer in reversed(range(0,self.number_of_layers)):
if layer == self.number_of_layers-1:
Rlt_state[layer],parD = self.call_Decoder(layer,Elt_state[layer],\
Rlt_top,Rlt_state[layer],epoch)
Rlt_top = Rlt_state[layer]
Rlt_state_tmp[layer] = Rlt_state[layer]
else:
Rlt_state[layer],parD = self.call_Decoder(layer,Elt_state[layer],\
Rlt_top,Rlt_state[layer],epoch)
Rlt_top = Rlt_state[layer]
Rlt_state_tmp[layer] = Rlt_state[layer]
# 2.a) Discriminative part:
for layer in range(0,self.number_of_layers):
if layer == 0:
Elt_state[layer],parG = self.call_Encoder(layer,x_t,\
Rlt_state[layer],True,epoch)
Elt_state_tmp[layer] = Elt_state[layer]
else:
Elt_state[layer],parG = self.call_Encoder(layer,Elt_state[layer-1],\
Rlt_state[layer],False,epoch)
Elt_state_tmp[layer] = Elt_state[layer]
pDlist.extend(parD)
pGlist.extend(parG)
return Elt_state_tmp,Rlt_state_tmp,\
pDlist,pGlist
## Util functions ##
def getResult(data):
return torch.chunk(data,2,1)[0][0]
def getOptimizer(par,lr):
optimizer = torch.optim.Adam(par.parameters(),\
lr=lr)
def saveImage(dir_path,ImageTg,itr):
ImageTg = np.asarray(ImageTg.data.cpu().numpy())
ImageTg = np.transpose(ImageTg, (2, 1, 0))
dataPathTg = os.path.join(dir_path,"Image_"+str(itr)+".jpg")
cv2.imwrite(dataPathTg,ImageTg)
####
def test():
#######################
image_size = 256
kernel_size = 3
num_layers = 6
max_epoch = 100
T = 6
lr = 0.1
SaveTrainImage = ""
#####################
SaveModelPath = ""
OpenModelPath = ""
numSaveIter = 0
######################
# This must always be kept at zero !!
y_lt = Variable(torch.zeros(1, 6, image_size, image_size)).cuda()
######################
cuda_flag = True
totLoss = 0
prednet_model = PrednetModel(num_layers,image_size,kernel_size,\
SaveModelPath,OpenModelPath,\
numSaveIter)
print('Create a MSE criterion')
criterion = nn.MSELoss()
criterion = criterion.cuda()
for epoch in range(0,max_epoch):
# init states:
Rlt_state = [None] * (num_layers)
Elt_state = [None] * (num_layers)
listError = []
# get seq of images from dataset, now is just a random seq !!
x_lt = Variable(torch.randn(T, 1, 3, image_size, image_size)).cuda()
# set tmp error value
E_lt = Variable(torch.randn(T, 1, 6, image_size, image_size)).cuda()
outImg = None
pDlist = 0
pGlist = 0
loss = 0
for t in range(0, T):
print("==> Time step: " + str(t))
Elt_state,Rlt_state,\
pDlist,pGlist = prednet_model(x_lt[t],Elt_state,Rlt_state,epoch)
# IMPLEMENT: equation number 5 of PredNet paper.
# Sum of MSE of seq time errors sum_t(error_t-0) !!
loss += criterion(Elt_state[0], y_lt)
outImg = Elt_state[0]
loss.backward()
# clip par on GRU !!
torch.nn.utils.clip_grad_norm(pDlist,0.5)
getOptimizer(pDlist,lr)
getOptimizer(pGlist,lr)
totLoss = loss.data[0]/max_epoch
print("==> Loss at epoch: [" + str(epoch) + "] is: " + str(totLoss))
#print("==> Save last train image:")
#image = getResult(outImg)
#saveImage(SaveTrainImage,image,epoch)
if __name__ == '__main__':
test()