-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathanalyze_reconstructions.py
325 lines (253 loc) · 15.6 KB
/
analyze_reconstructions.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
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
#possibly change no_folders
#change weights
#change fn folders
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import datasets, transforms
from torch.autograd import Variable
from os import listdir
from os.path import isfile, join
from scipy.io import loadmat # for loading mat files
import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1 import make_axes_locatable
import math
fr_pervid = 49 #19 #how many frames are in a video after convolution w/ filters has been done!
tau = 2 #1
N = 50 #20; %how many frames the video has
end_fr = 47 #18; %how many frames are there after looking in the past + doing filter (ie. how many frames does XTrain_Prep_forPython have?)
delta_t = 1 #how many frames the filter has-1?!
N_x = 167 #161
N_y = 167 #161
filters34_2 = loadmat('data_filts_px3_v2.mat')
filters34_temp = np.array(filters34_2['data_filts2'][0,0:16].tolist())
filters34_temp = np.expand_dims(filters34_temp, axis=0)
filters34_temp = np.transpose(filters34_temp, (0,1,4,2,3))
filters34_notemp = np.array(filters34_2['data_filts2'][0,16:34].tolist())
filters34_notemp = np.expand_dims(filters34_notemp, axis=1)
filters34_notemp = np.expand_dims(filters34_notemp, axis=0)
filters34_temp = np.transpose(np.transpose(filters34_temp, (0,2,3,4,1))-filters34_temp.reshape(1,filters34_temp.shape[1],-1).mean(axis=2), (0,4,1,2,3))
filters34_notemp = np.transpose(np.transpose(filters34_notemp, (0,2,3,4,1))-filters34_notemp.reshape(1,filters34_notemp.shape[1],-1).mean(axis=2), (0,4,1,2,3))
filter_len = np.shape(filters34_temp)[2]
NF_expl = [np.shape(filters34_temp)[1],np.shape(filters34_notemp)[1]]
NF = NF_expl[0] + NF_expl[1]
# 18filters, normal model, mov+stat
XTrain = np.load('/home/dvoina/simple_vids/repository_project1/XTrain_forPython_34filters_NoiseG2.5_reviewSimple.npy')
XTrain_real = np.load('/home/dvoina/simple_vids/repository_project1/XTrain_forPython_34filters_noNoise_reviewSimple.npy')
# 18filters, rnn model, approx
W_stat = np.load('/home/dvoina/simple_vids/results/W_43x43_34filters_static_simple_3px_ReviewSparse2.npy')
W_mov = np.load('/home/dvoina/simple_vids/results/W_43x43_34filters_moving_simple_3px_tau2_ReviewSparse.npy')
W_s2p = np.copy(W_stat); W_s2p[W_s2p>0] = 0
W_stat_matrix_torch = torch.from_numpy(W_stat).float()
W_mov_matrix_torch = torch.from_numpy(W_mov).float()
W_s2p_torch = torch.from_numpy(W_s2p).float()
Wp2v = np.load('/home/dvoina/simple_vids/results/Wp2v_34_param_simple3px_sst2vip_18_Nf5_review2.npy')
Wv2p = np.load('/home/dvoina/simple_vids/results/Wv2p_34_param_simple3px_sst2vip_18_Nf5_review2.npy')
Wv2s = np.load('/home/dvoina/simple_vids/results/Wv2s_34_param_simple3px_sst2vip_18_Nf5_review2.npy')
Wp2v = torch.from_numpy(np.squeeze(Wp2v[-3,2,:,:,:,:]))
Wv2p = torch.from_numpy(np.squeeze(Wv2p[-3,2,:,:,:,:]))
Wv2s = torch.from_numpy(np.squeeze(Wv2s[-3,2,:,:,:,:]))
dir_ref = '/home/dvoina/simple_vids/moving_videos_bsr_jumps_simple_3px_valid/'
#address = '/home/dvoina/simple_vids/moving_videos_bsr_jumps_simple_8px_test/'
address = '/home/dvoina/simple_vids/moving_videos_bsr_jumps_simple_3px_valid/'
folder_list = listdir(address)
no_folders = np.shape(folder_list)[0]
#######
c1 = 3; c2 = 3; c3 = 3;
n1 = int(math.floor(c1/2)); n2 = int(math.floor(c2/2)); n3 = int(math.floor(c3/2));
Nf = 34; Nf2 = 5
class Activities(torch.nn.Module):
def __init__(self):
super(Activities, self).__init__()
self.conv_p2v = torch.nn.Conv2d(Nf, Nf2, c1, bias=None, padding=0)
self.conv_v2s = torch.nn.Conv2d(Nf2, Nf, c2, bias=None, padding=0)
self.conv_s2p = torch.nn.Conv2d(Nf, Nf, 43, bias=None, padding=0) # conv for sure
self.conv_v2p = torch.nn.Conv2d(Nf2, Nf, c3, bias=None, padding=0)
self.conv_mov = torch.nn.Conv2d(Nf, Nf, 43, bias=None, padding=0) # conv for sure
self.conv_stat = torch.nn.Conv2d(Nf, Nf, 43, bias=None, padding=0)
def forward(self, x, x_prev):
x1 = self.conv_p2v(x_prev)
y1 = -self.conv_v2p(x1)
y2 = -self.conv_v2s(x1)
y3 = self.conv_s2p(y2)
xmov = self.conv_mov(x_prev)
xstat = self.conv_stat(x_prev)
if (n1+n3!=0):
xmov = xmov[:,:,n3+n1:-n3-n1, n3+n1:-n3-n1]
xstat = xstat[:,:,n3+n1:-n3-n1, n3+n1:-n3-n1]
y1 = y1[:,:,21:-21, 21:-21]
x = x[:,:,21+n3+n1:-21-n3-n1,21+n3+n1:-21-n3-n1]
ones = torch.ones(x.size()[0], x.size()[1], x.size()[2], x.size()[3]).cuda()
a_static = x*(ones + xstat)
a_moving = x*(ones + xmov)
xapprox = xstat + y1 + y3
a_approx = x*(ones + xapprox)
return a_static, a_moving, a_approx, xstat, xmov
#return x*xstat, x*xmov, x*xapprox
class inverse(torch.nn.Module):
def __init__(self):
super(inverse, self).__init__()
self.conv_inv = torch.nn.Conv2d(1, 1, 15, bias=None, padding=0)
def forward(self, x):
y = self.conv_inv(x)
return y
cuda = torch.cuda.is_available()
model = Activities()
model.conv_s2p.weight.data.copy_(W_s2p_torch)
model.conv_mov.weight.data.copy_(W_mov_matrix_torch)
model.conv_stat.weight.data.copy_(W_stat_matrix_torch)
model.conv_p2v.weight.data.copy_(Wp2v)
model.conv_v2p.weight.data.copy_(Wv2p)
model.conv_v2s.weight.data.copy_(Wv2s)
inv = inverse()
if cuda:
model.cuda()
inv.cuda()
def mean2(x):
y = np.sum(x) / np.size(x);
return y
def corr2(a,b):
a = a - mean2(a)
b = b - mean2(b)
r = (a*b).sum() / math.sqrt((a*a).sum() * (b*b).sum());
return r
#alpha_array = [0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1, 1.5,100]
#no_folders = 1
alpha_array = [1]
Rtot_moving = np.zeros((np.shape(alpha_array)[0], no_folders, end_fr-delta_t))
R_moving = np.zeros((np.shape(alpha_array)[0], no_folders))
Rtot_static = np.zeros((np.shape(alpha_array)[0], no_folders, end_fr-delta_t))
R_static = np.zeros((np.shape(alpha_array)[0], no_folders))
Rtot_real = np.zeros((np.shape(alpha_array)[0], no_folders, end_fr-delta_t))
R_real = np.zeros((np.shape(alpha_array)[0], no_folders))
Rtot_ff = np.zeros((np.shape(alpha_array)[0], no_folders, end_fr-delta_t))
R_ff = np.zeros((np.shape(alpha_array)[0], no_folders))
Rtot_approx = np.zeros((np.shape(alpha_array)[0], no_folders, end_fr-delta_t))
R_approx = np.zeros((np.shape(alpha_array)[0], no_folders))
A_static2 = np.zeros((no_folders, end_fr, NF, N_x-42-2*n1-2*n3, N_y-42-2*n1-2*n3))
A_moving2 = np.zeros((no_folders, end_fr, NF, N_x-42-2*n1-2*n3, N_y-42-2*n1-2*n3))
A_approx2 = np.zeros((no_folders, end_fr, NF, N_x-42-2*n1-2*n3, N_y-42-2*n1-2*n3))
A_static_max = np.zeros((no_folders))
A_moving_max = np.zeros((no_folders))
A_approx_max = np.zeros((no_folders))
#for alpha in [0.5, 0.75, 1, 1.25, 1.5,100]:
for index, alpha in enumerate(alpha_array):
print(alpha)
for i in range(no_folders):
#for j in range(1,2):
#i=j
print(i)
filename = folder_list[i][:-4]
XTrain_main = XTrain[i*fr_pervid+tau : i*fr_pervid + fr_pervid, 0, :, :, :]
XTrain_prev = XTrain[i*fr_pervid : i*fr_pervid + fr_pervid - tau, 0, :, :, :]
XTrain_main_real = XTrain_real[i * fr_pervid + tau: i * fr_pervid + fr_pervid, 0, :, :, :]
XTrain_main = torch.from_numpy(XTrain_main).squeeze(1)
XTrain_prev = torch.from_numpy(XTrain_prev).squeeze(1)
XTrain_main_real = torch.from_numpy(XTrain_main_real).squeeze(1)
if cuda:
x = XTrain_main.cuda()
x_prev = XTrain_prev.cuda()
A_static, A_moving, A_approx, xstat, xmov = model(x, x_prev)
A_static = A_static.cpu().detach().numpy(); A_moving = A_moving.cpu().detach().numpy(); A_approx = A_approx.cpu().detach().numpy();
A_static[A_static<0] = 0; A_moving[A_moving<0] = 0; A_approx[A_approx<0] = 0;
A_static_max[i] = np.max(xstat.cpu().detach().numpy()); A_moving_max[i] = np.max(xmov.cpu().detach().numpy()); A_approx_max[i] = np.max(A_approx);
Recon_moving_initial = np.zeros((np.shape(A_moving)[0]+1, np.shape(A_moving)[1], XTrain_main.size()[2] - 42-2*n1-2*n3, XTrain_main.size()[3] - 42-2*n1-2*n3))
Recon_static_initial = np.zeros((np.shape(A_static)[0]+1, np.shape(A_static)[1], XTrain_main.size()[2] - 42-2*n1-2*n3, XTrain_main.size()[3] - 42-2*n1-2*n3))
Recon_approx_initial = np.zeros((np.shape(A_approx)[0]+1, np.shape(A_approx)[1], XTrain_main.size()[2] - 42-2*n1-2*n3, XTrain_main.size()[3] - 42-2*n1-2*n3))
Recon_real_initial = np.zeros((np.shape(XTrain_main_real)[0]+1, np.shape(XTrain_main_real)[1], np.shape(XTrain_main_real)[2] - 42-2*n1-2*n3, np.shape(XTrain_main_real)[3] - 42-2*n1-2*n3))
Recon_ff_initial = np.zeros((np.shape(XTrain_main)[0]+1, np.shape(XTrain_main)[1], np.shape(XTrain_main)[2] - 42-2*n1-2*n3, np.shape(XTrain_main)[3] - 42-2*n1-2*n3))
for frame in range(end_fr):
for f in range(NF):
if f < NF_expl[0]:
inv_filt = torch.from_numpy(np.fliplr(np.flipud(filters34_temp[0,f,0,:,:])).copy()).unsqueeze(0).unsqueeze(0)
inv.conv_inv.weight.data.copy_(inv_filt.cuda())
activity = torch.from_numpy(np.pad(XTrain_main[frame, f, 21+n1+n3:-21-n1-n3, 21+n1+n3:-21-n1-n3], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_ff_initial[frame, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(A_moving[frame,f,:,:], (7,7), 'symmetric')).unsqueeze(0).unsqueeze(0).float()
Recon_moving_initial[frame,f,:,:] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(A_static[frame, f, :, :], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_static_initial[frame, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(A_approx[frame, f, :, :], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_approx_initial[frame, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(XTrain_main_real[frame, f, 21+n1+n3:-21-n1-n3, 21+n1+n3:-21-n1-n3], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_real_initial[frame, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
else:
inv_filt = torch.from_numpy(np.fliplr(np.flipud(filters34_notemp[0, f - NF_expl[0], 0, :, :])).copy()).unsqueeze(0).unsqueeze(0)
inv.conv_inv.weight.data.copy_(inv_filt.cuda())
activity = torch.from_numpy(np.pad(XTrain_main[frame, f, 21+n1+n3:-21-n1-n3, 21+n1+n3:-21-n1-n3], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_ff_initial[frame+delta_t, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(A_moving[frame, f, :, :], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0).float()
Recon_moving_initial[frame+delta_t,f,:,:] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(A_static[frame, f, :, :], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_static_initial[frame+delta_t, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(A_approx[frame, f, :, :], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_approx_initial[frame+delta_t, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
activity = torch.from_numpy(np.pad(XTrain_main_real[frame, f, 21+n1+n3:-21-n1-n3, 21+n1+n3:-21-n1-n3], (7, 7), 'symmetric')).unsqueeze(0).unsqueeze(0)
Recon_real_initial[frame+delta_t, f, :, :] = inv(activity.cuda()).cpu().detach().numpy()
Recon_moving_final = np.mean(Recon_moving_initial[delta_t:-delta_t,:,:,:], axis = 1)
#Recon_moving_final = np.sum(Recon_moving_initial[delta_t:-delta_t, :, :, :]*weights[np.newaxis, np.newaxis, :, np.newaxis, np.newaxis], axis=2)
Recon_moving_final = np.squeeze(Recon_moving_final)
Recon_static_final = np.mean(Recon_static_initial[delta_t:-delta_t, :, :, :], axis=1)
#Recon_static_final = np.sum(Recon_static_initial[delta_t:-delta_t, :, :, :]*weights[np.newaxis, np.newaxis, :, np.newaxis, np.newaxis], axis=2)
Recon_static_final = np.squeeze(Recon_static_final)
Recon_approx_final = np.mean(Recon_approx_initial[delta_t:-delta_t, :, :, :], axis=1)
Recon_approx_final = np.squeeze(Recon_approx_final)
Recon_real_final = np.mean(Recon_real_initial[delta_t:-delta_t, :, :, :], axis=1)
#Recon_real_final = np.sum(Recon_real_initial[delta_t:-delta_t, :, :, :]*weights[np.newaxis, np.newaxis, :, np.newaxis, np.newaxis], axis=2)
Recon_real_final = np.squeeze(Recon_real_final)
Recon_ff_final = np.mean(Recon_ff_initial[delta_t:-delta_t, :, :, :], axis=1)
#Recon_ff_final = np.sum(Recon_ff_initial[delta_t:-delta_t, :, :, :]*weights[np.newaxis, np.newaxis, :, np.newaxis, np.newaxis], axis=2)
Recon_ff_final = np.squeeze(Recon_ff_final)
vid = loadmat(dir_ref+filename)
video = vid['s_modified']
r_ff = np.zeros((1,end_fr-delta_t))
r_static = np.zeros((1,end_fr-delta_t))
r_moving = np.zeros((1, end_fr - delta_t))
r_real = np.zeros((1, end_fr-delta_t))
r_approx = np.zeros((1, end_fr-delta_t))
step = N - end_fr
for frame in range(end_fr-delta_t):
img = np.squeeze(video[frame+step, :, :])
img = img - np.mean(img)
if np.max(img) != 0:
img = img/np.max(img)
img = img[21+n1+n3:-21-n1-n3, 21+n1+n3:-21-n1-n3]
#r_moving[0,frame] = corr2(img, np.squeeze(Recon_moving_final[frame,:,:]))
#r_static[0,frame] = corr2(img, np.squeeze(Recon_static_final[frame, :, :]))
#r_real[0,frame] = corr2(img, Recon_real_final[frame, :, :])
#r_ff[0,frame] = corr2(img, Recon_ff_final[frame, :, :])
#r_approx[0,frame] = corr2(img, Recon_approx_final[frame,:,:])
r_ff[0, frame] = corr2(Recon_real_final[frame, :, :], Recon_ff_final[frame, :, :])
r_moving[0, frame] = corr2(Recon_real_final[frame, :, :], Recon_moving_final[frame, :, :])
r_static[0, frame] = corr2(Recon_real_final[frame, :, :], Recon_static_final[frame, :, :])
r_approx[0, frame] = corr2(Recon_real_final[frame, :, :], Recon_approx_final[frame, :, :])
#i = 0
A_static2[i,:,:,:,:] = A_static
A_moving2[i,:,:,:,:] = A_moving
A_approx2[i,:,:,:,:] = A_approx
Rtot_moving[index, i, :] = r_moving
R_moving[index, i] = np.mean(r_moving)
Rtot_static[index, i, :] = r_static
R_static[index, i] = np.mean(r_static)
Rtot_approx[index, i, :] = r_approx
R_approx[index, i] = np.mean(r_approx)
#Rtot_real[index, i, :] = r_real
#R_real[index, i] = np.mean(r_real)
Rtot_ff[index, i] = r_ff
R_ff[index, i] = np.mean(r_ff)
Rtot_approx[index, i] = r_approx
R_approx[index, i] = np.mean(r_approx)
print(np.mean(R_ff[0,:]))
print(np.mean(R_static[0,:]))
print(np.mean(R_moving[0,:]))
print(np.mean(R_approx[0,:]))
Rtot = np.concatenate((Rtot_ff, Rtot_static, Rtot_moving, Rtot_approx), axis = 0)
np.save('Rtot_34filters_3px_G05_tau2i_'+filename+'.npy', Rtot)
Recon_real = Recon_real_final[np.newaxis,:,:,:]
Recon_ff = Recon_ff_final[np.newaxis,:,:,:]
Recon_static = Recon_static_final[np.newaxis,:,:,:]
Recon_moving = Recon_moving_final[np.newaxis,:,:,:]
Recon_approx = Recon_approx_final[np.newaxis,:,:,:]
Recon = np.concatenate((Recon_real, Recon_ff, Recon_static, Recon_moving, Recon_approx), axis = 0)
np.save('Recon'+filename+'_tau2i_G05_34.npy', Recon)