-
Notifications
You must be signed in to change notification settings - Fork 0
/
mp4-test.py
279 lines (245 loc) · 11.1 KB
/
mp4-test.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
import imageio
import sys
import multiprocessing
import numpy as np
import matplotlib.pyplot as plt
from scipy import ndimage
from scipy.ndimage.measurements import center_of_mass
from scipy.ndimage import zoom
from scipy.ndimage import geometric_transform
from scipy.interpolate import griddata
from scipy.ndimage.interpolation import map_coordinates
from mad import median_absolute_deviation
def trans_worker(params):
frame, average, patch_size, crop = params
grid_edges = np.arange(0,2*crop+1,patch_size)
grid_x, grid_y = np.mgrid[0:2*crop-1:np.complex(2*crop), 0:2*crop-1:np.complex(2*crop)]
source = []
destination = []
xx_out = []
yy_out = []
for x in xrange(len(grid_edges)):
for y in xrange(len(grid_edges)):
destination.append([grid_edges[x],grid_edges[y]])
x_0 = grid_edges[x]
y_0 = grid_edges[y]
x_m = patch_size/2 if x > 0 else 0
y_m = patch_size/2 if y > 0 else 0
x_p = patch_size/2 if x < len(grid_edges)-1 else 0
y_p = patch_size/2 if y < len(grid_edges)-1 else 0
x_low = x_0 - x_m
x_high = x_0 + x_p
y_low = y_0 - y_m
y_high = y_0 + y_p
patch = frame[x_low:x_high,y_low:y_high]
res = []
for move_x in xrange(-x_m,x_p):
for move_y in xrange(-y_m,y_p):
ref_x0 = x_low+move_x
ref_x1 = x_high+move_x
ref_y0 = y_low+move_y
ref_y1 = y_high+move_y
ref_patch = average[ref_x0:ref_x1,ref_y0:ref_y1]
res.append(np.sum(np.square(patch-ref_patch)))
minimum = np.argmin(res)
x_min = minimum/(x_m+x_p)-x_m
y_min = minimum%(y_m+y_p)-y_m
x_out = grid_edges[x]-x_min
y_out = grid_edges[y]-y_min
xx_out.append(x_out)
yy_out.append(y_out)
source.append([x_out, y_out])
source = np.array(source)
destination = np.array(destination)
grid_z = griddata(destination, source, (grid_x, grid_y), method='cubic')
trans = []
for grid_z_p in grid_z:
trans.append(map_coordinates(frame, grid_z_p.T,mode='reflect'))
return trans
class CIA_enhance(object):
def __init__(self, filename, n_frames=None, patch_size=8):
self.reader = imageio.get_reader(filename, 'ffmpeg')
self.fps = self.reader.get_meta_data()['fps']
vid_shape = self.reader.get_data(0).shape[:2]
self.vid_shape = np.array(vid_shape)
self.center = self.vid_shape/2
if n_frames:
assert(n_frames < self.reader.get_length())
self.n_frames = n_frames
else:
self.n_frames = self.reader.get_length() - 1
#self.crop = 256
#self.crop = 160
self.crop = 512
assert(self.crop % patch_size == 0)
self.patch_size = patch_size
def align_and_crop(self):
print '--- coarse alignement ---'
self.seq = []
align_seq = []
self.alignement_error = [[],[]]
for i in xrange(0,self.n_frames):
# greyscale conversion
img = np.average(self.reader.get_data(i),axis=2)
for n in range(2):
com = ndimage.measurements.center_of_mass(img)
for axis in [0,1]:
img = np.roll(img, int(round(self.center[axis]-com[axis])), axis=axis)
com = np.array(ndimage.measurements.center_of_mass(img))
self.alignement_error[n].append(np.sum(np.square(com-self.center)))
align_seq.append(img)
#fine align
print '--- fine alignement ---'
lower_bounds = self.center - self.crop
upper_bounds = self.center + self.crop
self.seq.append(align_seq[0][lower_bounds[0]:upper_bounds[0],lower_bounds[1]:upper_bounds[1]])
for frame in align_seq[1:]:
res = []
for move_x in range(-10,10):
for move_y in range(-10,10):
test = frame[lower_bounds[0]+move_x:upper_bounds[0]+move_x,lower_bounds[1]+move_y:upper_bounds[1]+move_y]
res.append(np.sum(np.square(test-self.seq[0])))
minimum = np.argmin(res)
x_min = minimum/(20)-10
y_min = minimum%(20)-10
self.seq.append(align_seq[0][lower_bounds[0]+x_min:upper_bounds[0]+x_min:,lower_bounds[1]+y_min:upper_bounds[1]+y_min])
def transform(self):
print '--- image warping ---'
self.seq = np.array(self.seq)
self.average = np.average(self.seq,axis=0)
#self.seq_trans = []
average_list = [self.average]*len(self.seq)
params = zip(self.seq, average_list, [self.patch_size]*len(self.seq) , [self.crop]*len(self.seq))
pool = multiprocessing.Pool()
self.seq_trans = pool.map(trans_worker, params)
#for frame in self.seq:
# trans = trans_worker(frame, average, self.patch_size, self.crop)
# self.seq_trans.append(np.array(trans))
self.seq_trans = np.array(self.seq_trans)
average_trans = np.average(self.seq_trans,axis=0)
average_list = np.array(average_list)
residuals_before = np.mean(np.square(self.seq - average_list),axis=(1,2))
residuals_after = np.mean(np.square(self.seq_trans - average_list),axis=(1,2))
central_residuals_before = np.mean(np.square(self.seq - average_list)[:,2*self.patch_size:-2*self.patch_size,2*self.patch_size:-2*self.patch_size],axis=(1,2))
central_residuals_after = np.mean(np.square(self.seq_trans - average_list)[:,2*self.patch_size:-2*self.patch_size,2*self.patch_size:-2*self.patch_size],axis=(1,2))
#fig = plt.figure(figsize=(15,15), frameon=False)
#fig.subplots_adjust(hspace=0)
#fig.subplots_adjust(wspace=0)
#ax1 = fig.add_subplot(1, 2, 1)
#ax1.imshow(average, cmap='Greys_r')
#ax1.set_xlim(grid_edges[0],grid_edges[-1])
#ax1.set_ylim(grid_edges[0],grid_edges[-1])
#ax1.set_title('Average')
#ax2 = fig.add_subplot(1, 2, 2)
#ax2.imshow(average_trans, cmap='Greys_r')
#ax2.set_xlim(grid_edges[0],grid_edges[-1])
#ax2.set_ylim(grid_edges[0],grid_edges[-1])
#ax2.set_title('Average Transformed')
#writer = imageio.get_writer('average.jpeg',quality=100)
#writer.append_data(average)
#writer.close()
#writer = imageio.get_writer('average.tif')
#writer.append_data(average)
#writer.close()
#writer = imageio.get_writer('taverage.jpeg',quality=100)
#writer.append_data(average_trans)
#writer.close()
#writer = imageio.get_writer('taverage.tif')
#writer.append_data(average_trans)
#writer.close()
#i = 3
#fig = plt.figure(figsize=(20,12), frameon=False)
#fig.subplots_adjust(hspace=0)
#fig.subplots_adjust(wspace=0)
#ax1 = fig.add_subplot(2, 3, 1)
#ax1.imshow(average, cmap='Greys_r')
#ax1.set_title('Average')
##ax1.contour(average)
#ax6 = fig.add_subplot(2,3,4)
##ax6.plot(residuals_before,color='r',linestyle=':')
##ax6.plot(residuals_after, color='b',linestyle=':')
##ax6.plot(central_residuals_before,color='r')
#ax6.plot(central_residuals_after,color='b')
#ax6.plot(self.alignement_error[1],color='g')
##ax6.axhline(np.average(residuals_before),color='r',linestyle=':')
##ax6.axhline(np.average(residuals_after),color='b',linestyle=':')
##ax6.axhline(np.average(central_residuals_before),color='r')
#ax6.axhline(np.average(central_residuals_after),color='b')
#ax6.axhline(np.average(self.alignement_error[1]),color='g')
#ax2 = fig.add_subplot(2, 3, 2)
#ax2.imshow(self.seq[i], cmap='Greys_r')
#ax2.set_title('Frame %i'%i)
#ax4 = fig.add_subplot(2, 3, 5)
#im = ax4.imshow(np.square(self.seq[i]-average),vmin=0, vmax=20)
#ax4.set_title('Residuals')
#plt.colorbar(im,orientation="horizontal")
#ax3 = fig.add_subplot(2, 3, 3)
#ax3.imshow(self.seq_trans[i],cmap='Greys_r')
#ax3.set_title('Transform')
#ax5 = fig.add_subplot(2, 3, 6)
#im = ax5.imshow(np.square(self.seq_trans[i]-average),vmin=0, vmax=20)
#plt.colorbar(im,orientation="horizontal")
#ax5.set_title('Residuals')
#fig.set_tight_layout(True)
#plt.show()
def temporal_kernel_regression(self):
print '--- assembly ---'
new = np.zeros_like(self.seq_trans[0])
# smoothing param
mu = 1.
for x in xrange(new.shape[0]):
for y in xrange(new.shape[1]):
center_x = self.patch_size/2 if x > self.patch_size/2 else x
center_y = self.patch_size/2 if y > self.patch_size/2 else y
lower_bounds_x = x - center_x
upper_bounds_x = x + self.patch_size/2 - 1 if x+self.patch_size/2 <= new.shape[0] else new.shape[0]
lower_bounds_y = y - center_y
upper_bounds_y = y + self.patch_size/2 - 1 if y+self.patch_size/2 <= new.shape[1] else new.shape[1]
var = []
patches = self.seq_trans[:,lower_bounds_x:upper_bounds_x,lower_bounds_y:upper_bounds_y]
for patch in patches:
variance = np.sum(np.square(patch - np.mean(patch)))
L2 = patch.shape[0]*patch.shape[1]
var.append(variance*1./(L2-1))
ref_patch = patches[np.argmax(var)]
Us = []
Urs = []
for patch in patches:
U = np.sum(np.square(patch - ref_patch))
U /= L2
sigma_n2 = median_absolute_deviation(patch)
U -= 2*sigma_n2
Ux = np.exp(-U/(mu**2))
Us.append(Ux)
Urs.append(Ux*patch[center_x,center_y])
Us = np.array(Us)
Urs = np.array(Urs)
new[x,y] = np.sum(Urs)/np.sum(Us)
writer = imageio.get_writer('diflim.jpeg',quality=100)
writer.append_data(new)
writer.close()
writer = imageio.get_writer('diflim.tif')
writer.append_data(new)
writer.close()
fig = plt.figure(figsize=(20,8), frameon=False)
fig.subplots_adjust(hspace=0)
fig.subplots_adjust(wspace=0)
ax1 = fig.add_subplot(1, 3, 1)
ax1.imshow(self.average, cmap='Greys_r',interpolation='nearest')
ax1.set_title('average')
ax2 = fig.add_subplot(1, 3, 2)
ax2.imshow(new, cmap='Greys_r',interpolation='nearest')
ax2.set_title('new')
ax3 = fig.add_subplot(1, 3, 3)
ax3.plot(var)
fig.set_tight_layout(True)
plt.show()
if __name__ == '__main__':
filename = 'moon-00002.mp4'
#filename = 'oaCapture-20150306-202356.avi'
#filename = 'saturn-20150605-233034-000000.avi'
#filename = 'moon_close.avi'
enhancer = CIA_enhance(filename,160,patch_size=16)
enhancer.align_and_crop()
enhancer.transform()
enhancer.temporal_kernel_regression()