forked from BVLC/caffe
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathvisualize_results.py
303 lines (265 loc) · 12.8 KB
/
visualize_results.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
import caffe
import matplotlib.pylab as plt
import numpy as np
import h5py
import copy
def vis_square_color_pts(mydata,pts,timesteps,filterind=[],redpoints=[], padsize=1, padval=0,alpha=1.0, filename=[]):
ind = timesteps[0]
if len(timesteps) == 1:
data = copy.deepcopy(mydata[ind:ind+1,:,:,:].transpose(0,2,3,1))
else:
if len(filterind) > 0:
print "WARNING: points may not be correct"
data = copy.deepcopy(mydata[np.sort(timesteps),:,:,:].transpose(0,2,3,1))
#for i in range(data.shape[0]):
# data[i,:,:,:] -= data[i,:,:,:].max()
# data[i,:,:,:] /= data[i,:,:,:].min()
data -= data.min()
data /= data.max()
img_width = mydata.shape[2]
# force the number of filters to be square
n = int(np.ceil(np.sqrt(data.shape[0])))
padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))
# tile the filters into an image
data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
plt.imshow(data, alpha = alpha)
plt.axis("off")
if len(filterind) != 0:
ax=plt.gca();
NUM_COLORS=len(filterind)
cm = plt.get_cmap('nipy_spectral');
redi = []
#ax.set_color_cycle([cm(1.*i/NUM_COLORS) for i in range(NUM_COLORS)]);
for i in range(len(filterind)):
filteri = filterind[i]
if filteri in redpoints:
redi.append(filteri);
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'or', markersize=10);
else:
#plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'o', color=cm((1.*i)/NUM_COLORS),markersize=7);
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'ob', markersize=10,markeredgecolor=cm((1.*i)/NUM_COLORS),fillstyle='none', markeredgewidth=3.0);
for filteri in redi:
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'or', markersize=10);
plt.xlim([0,img_width]);plt.ylim([img_width,0]);
if filename: plt.savefig(filename, bbox_inches='tight')
def vis_square_color_pts_compare(mydata,pts,pts2,timesteps,filterind=[],redpoints=[], padsize=1, padval=0,alpha=1.0, filename=[]):
ind = timesteps[0]
if len(timesteps) == 1:
data = copy.deepcopy(mydata[ind:ind+1,:,:,:].transpose(0,2,3,1))
else:
if len(filterind) > 0:
print "WARNING: points may not be correct"
data = copy.deepcopy(mydata[np.sort(timesteps),:,:,:].transpose(0,2,3,1))
#for i in range(data.shape[0]):
# data[i,:,:,:] -= data[i,:,:,:].max()
# data[i,:,:,:] /= data[i,:,:,:].min()
data -= data.min()
data /= data.max()
img_width = mydata.shape[2]
# force the number of filters to be square
n = int(np.ceil(np.sqrt(data.shape[0])))
padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))
# tile the filters into an image
data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
plt.imshow(data, alpha = alpha)
plt.axis("off")
if len(filterind) != 0:
ax=plt.gca();
NUM_COLORS=len(filterind)
cm = plt.get_cmap('nipy_spectral');
redi = []
#ax.set_color_cycle([cm(1.*i/NUM_COLORS) for i in range(NUM_COLORS)]);
for i in range(len(filterind)):
filteri = filterind[i]
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'x', markersize=12,markeredgecolor=(.894,0,0,1.0),fillstyle='none', markeredgewidth=3.0);
plt.plot((pts2[ind,filteri*2]+1)/2.0*img_width,(pts2[ind,filteri*2+1]+1)/2.0*img_width,'ob', markersize=12,markeredgecolor=(0,0,.894,0.7),fillstyle='none', markeredgewidth=3.0);
#plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'o', color=cm((1.*i)/NUM_COLORS),markersize=7);
#plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'ob', markersize=10);
for filteri in redi:
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'or', markersize=10);
plt.xlim([0,img_width]);plt.ylim([img_width,0]);
if filename: plt.savefig(filename, bbox_inches='tight')
def vis_square_color_pts_compare_discrep(mydata,pts,pts2,timesteps,timesteps2,filterind=[],redpoints=[], padsize=1, padval=0,alpha=1.0, filename=[]):
ind = timesteps[0]
ind2 = timesteps2[0]
if len(timesteps) == 1:
data = copy.deepcopy(mydata[ind:ind+1,:,:,:].transpose(0,2,3,1))
else:
if len(filterind) > 0:
print "WARNING: points may not be correct"
data = copy.deepcopy(mydata[np.sort(timesteps),:,:,:].transpose(0,2,3,1))
#for i in range(data.shape[0]):
# data[i,:,:,:] -= data[i,:,:,:].max()
# data[i,:,:,:] /= data[i,:,:,:].min()
data -= data.min()
data /= data.max()
img_width = mydata.shape[2]
# force the number of filters to be square
n = int(np.ceil(np.sqrt(data.shape[0])))
padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))
# tile the filters into an image
data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
plt.imshow(data, alpha = alpha)
plt.axis("off")
if len(filterind) != 0:
ax=plt.gca();
NUM_COLORS=len(filterind)
cm = plt.get_cmap('nipy_spectral');
redi = []
#ax.set_color_cycle([cm(1.*i/NUM_COLORS) for i in range(NUM_COLORS)]);
for i in range(len(filterind)):
filteri = filterind[i]
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'x', markersize=12,markeredgecolor=(.894,0,0,1.0),fillstyle='none', markeredgewidth=3.0);
plt.plot((pts2[ind2,filteri*2]+1)/2.0*img_width,(pts2[ind2,filteri*2+1]+1)/2.0*img_width,'ob', markersize=12,markeredgecolor=(0,0,.894,0.7),fillstyle='none', markeredgewidth=3.0);
#plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'o', color=cm((1.*i)/NUM_COLORS),markersize=7);
#plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'ob', markersize=10);
for filteri in redi:
plt.plot((pts[ind,filteri*2]+1)/2.0*img_width,(pts[ind,filteri*2+1]+1)/2.0*img_width,'or', markersize=10);
plt.xlim([0,img_width]);plt.ylim([img_width,0]);
if filename: plt.savefig(filename, bbox_inches='tight')
def vis_square_pts(mydata,pts,timesteps,filterinds,pointinds=[], padsize=1, padval=1, filename=[]):
if len(timesteps) == 1:
assert(len(timesteps) == 1)
t = timesteps[0];
data = -copy.deepcopy(mydata[t:t+1,filterinds,:,:].transpose(1,2,3,0))
else:
assert(len(filterinds) == 1)
i = filterinds[0]
data = -copy.deepcopy(mydata[timesteps,i:i+1,:,:].transpose(0,2,3,1))
for i in range(data.shape[0]):
data[i,:,:,:] -= data[i,:,:,:].min()
data[i,:,:,:] /= data[i,:,:,:].max()
# data -= data.min()
# data /= data.max()
img_width = mydata.shape[2]
print data.max()
print data.min()
# force the number of filters to be square
n = int(np.ceil(np.sqrt(data.shape[0])))
padding = ((0, n ** 2 - data.shape[0]), (0, padsize), (0, padsize)) + ((0, 0),) * (data.ndim - 3)
data = np.pad(data, padding, mode='constant', constant_values=(padval, padval))
# tile the filters into an image
data = data.reshape((n, n) + data.shape[1:]).transpose((0, 2, 1, 3) + tuple(range(4, data.ndim + 1)))
data = data.reshape((n * data.shape[1], n * data.shape[3]) + data.shape[4:])
plt.imshow(data[:,:,0], cmap='Greys', interpolation='nearest')
plt.axis("off")
if len(pointinds) != 0:
assert(len(timesteps) == 1)
t = timesteps[0];
#ax=plt.gca();
#NUM_COLORS=len(pointinds)
#cm = plt.get_cmap('nipy_spectral');
#ax.set_color_cycle([cm(1.*i/NUM_COLORS) for i in range(NUM_COLORS)]);
for i in range(len(pointinds)):
filteri = pointinds[i]
plt.plot((pts[t,filteri*2]+1)/2.0*img_width,(pts[t,filteri*2+1]+1)/2.0*img_width,'or');
plt.xlim([0,img_width]);plt.ylim([img_width,0]);
if filename: plt.savefig(filename, bbox_inches='tight')
class PoseResults:
def __init__(self, suffix):
plt.rcParams['figure.figsize'] = (10, 10)
prefix = "test_output/"
#prefix = "pose_baseline/"
#pose_output = h5py.File(prefix + suffix + ".h5", "r");
#self.pose = pose_output['data']
#self.pose_gt = pose_output['label']
softmax_output = h5py.File(prefix +"softmax_" + suffix + ".h5", "r");
self.softmax = softmax_output['data']
self.points = softmax_output['label']
conv1_output = h5py.File(prefix +"conv1_"+ suffix + ".h5", "r");
conv2_output = h5py.File(prefix +"conv2_"+ suffix + ".h5", "r");
conv3_output = h5py.File(prefix +"conv3_"+ suffix + ".h5", "r");
self.conv1 = conv1_output['data']
self.conv2 = conv2_output['data']
self.conv3 = conv3_output['data']
self.rgb = conv3_output['label']
#self.get_mean_dist()
#self.order_timesteps()
# defines the distances, and sorts everything in decreasing order of accuracy
def get_mean_dist(self, pt_idx=0):
diff = np.asarray(self.pose)-np.asarray(self.pose_gt);
diff = diff[:,pt_idx*3:pt_idx*3+3,0,0]
diffsq = np.power(diff,2);
eucsq = np.sum(diffsq,1);
print 'network output: ', np.mean(eucsq)/2
euc = np.power(eucsq,0.5);
loss=np.mean(euc);
std_dist = np.std(euc);
print 'mean distance: ', loss
print 'std distance: ', std_dist
self.euc = euc;
def order_timesteps(self):
order = np.argsort(self.euc)
self.euc = np.asarray([self.euc[i] for i in order])
self.rgb = np.asarray([self.rgb[i,:,:,:] for i in order])
self.conv1 = np.asarray([self.conv1[i,:,:,:] for i in order])
self.conv2 = np.asarray([self.conv2[i,:,:,:] for i in order])
self.conv3 = np.asarray([self.conv3[i,:,:,:] for i in order])
self.softmax = np.asarray([self.softmax[i,:,:,:] for i in order])
self.points = np.asarray([self.points[i,:,0,0] for i in order])
# Returns mean and standard deviation in distance for each point and all points together.
def get_mean_dists(self):
all_eucs = []
for pt_idx in range(3):
diff = np.asarray(self.pose)-np.asarray(self.pose_gt);
diff = diff[:,pt_idx*3:pt_idx*3+3,0,0]
diffsq = np.power(diff,2);
eucsq = np.sum(diffsq,1);
print 'network output: ', np.mean(eucsq)/2
euc = np.power(eucsq,0.5);
all_eucs.extend(euc)
loss=np.mean(euc);
std_dist = np.std(euc);
print 'mean distance: ', loss
print 'std distance: ', std_dist
self.euc = euc;
std_dist = np.std(all_eucs);
self.all_eucs = all_eucs
print 'all mean = ', np.mean(all_eucs)
print 'all std = ', np.std(all_eucs)
def vis_rgb(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
vis_square_color_pts(self.rgb, self.points, timesteps, filters,padsize=padsize, padval=padval);
def vis_conv1(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
if filters == []:
filters = [0]
vis_square_pts(self.conv1, self.points, timesteps, filters,padsize=padsize, padval=padval);
def vis_conv2(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
if filters == []:
filters = [0]
vis_square_pts(self.conv2, self.points, timesteps, filters,padsize=padsize, padval=padval);
def vis_conv3(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
if filters == []:
filters = [0]
vis_square_pts(self.conv3, self.points, timesteps, filters, padsize=padsize, padval=padval);
def vis_softmax(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
if filters == []:
filters = [0]
vis_square_pts(self.softmax, self.points, timesteps, filters,padsize=padsize, padval=padval);
def vis_conv3_pts(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
if filters == []:
filters = [0]
vis_square_pts(self.conv3, self.points, timesteps, filters, filters,padsize=padsize, padval=padval);
def vis_softmax_pts(self, timesteps = [], filters = [],padsize=1, padval=0):
if timesteps == []:
timesteps = [0];
if filters == []:
filters = [0]
vis_square_pts(self.softmax, self.points, timesteps, filters, filters,padsize=padsize, padval=padval);