forked from ankush-me/SynthText
-
Notifications
You must be signed in to change notification settings - Fork 1
/
synth_utils.py
282 lines (245 loc) · 9.08 KB
/
synth_utils.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
from __future__ import division
import numpy as np
from ransac import fit_plane_ransac
from sys import modules
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
#import mayavi.mlab as mym
class LUT_RGB(object):
"""
RGB LUT for Mayavi glyphs.
"""
def __create_8bit_rgb_lut__():
xl = np.mgrid[0:256, 0:256, 0:256]
lut = np.vstack((xl[0].reshape(1, 256**3),
xl[1].reshape(1, 256**3),
xl[2].reshape(1, 256**3),
255 * np.ones((1, 256**3)))).T
return lut.astype('int32')
__lut__ = __create_8bit_rgb_lut__()
@staticmethod
def rgb2scalar(rgb):
"""
return index of RGB colors into the LUT table.
rgb : nx3 array
"""
return 256**2*rgb[:,0] + 256*rgb[:,1] + rgb[:,2]
@staticmethod
def set_rgb_lut(myv_glyph):
"""
Sets the LUT of the Mayavi glyph to RGB-LUT.
"""
lut = myv_glyph.module_manager.scalar_lut_manager.lut
lut._vtk_obj.SetTableRange(0, LUT_RGB.__lut__.shape[0])
lut.number_of_colors = LUT_RGB.__lut__.shape[0]
lut.table = LUT_RGB.__lut__
def plot_xyzrgb(xyz,rgb,show=False):
"""
xyz : nx3 float
rgb : nx3 uint8
Plots a RGB-D point-cloud in mayavi.
"""
rgb_s = LUT_RGB.rgb2scalar(rgb)
pts_glyph = mym.points3d(xyz[:,0],xyz[:,1],xyz[:,2],
rgb_s,mode='point')
LUT_RGB.set_rgb_lut(pts_glyph)
if show:
mym.view(180,180)
mym.orientation_axes()
mym.show()
def visualize_plane(pt,plane,show=False):
"""
Visualize the RANSAC PLANE (4-tuple) fit to PT (nx3 array).
Also draws teh normal.
"""
# # plot the point-cloud:
if show and mym.gcf():
mym.clf()
# plot the plane:
plane_eq = '%f*x+%f*y+%f*z+%f'%tuple(plane.tolist())
m,M = np.percentile(pt,[10,90],axis=0)
implicit_plot(plane_eq, (m[0],M[0],m[1],M[1],m[2],M[2]))
# plot surface normal:
mu = np.percentile(pt,50,axis=0)
mu_z = -(mu[0]*plane[0]+mu[1]*plane[1]+plane[3])/plane[2]
mym.quiver3d(mu[0],mu[1],mu_z,plane[0],plane[1],plane[2],scale_factor=0.3)
if show:
mym.view(180,180)
mym.orientation_axes()
mym.show(True)
def implicit_plot(expr, ext_grid, Nx=11, Ny=11, Nz=11,
col_isurf=(50/255, 199/255, 152/255)):
"""
Function to plot algebraic surfaces described by implicit equations in Mayavi
Implicit functions are functions of the form
`F(x,y,z) = c`
where `c` is an arbitrary constant.
Parameters
----------
expr : string
The expression `F(x,y,z) - c`; e.g. to plot a unit sphere,
the `expr` will be `x**2 + y**2 + z**2 - 1`
ext_grid : 6-tuple
Tuple denoting the range of `x`, `y` and `z` for grid; it has the
form - (xmin, xmax, ymin, ymax, zmin, zmax)
fig_handle : figure handle (optional)
If a mayavi figure object is passed, then the surface shall be added
to the scene in the given figure. Then, it is the responsibility of
the calling function to call mlab.show().
Nx, Ny, Nz : Integers (optional, preferably odd integers)
Number of points along each axis. It is recommended to use odd numbers
to ensure the calculation of the function at the origin.
"""
xl, xr, yl, yr, zl, zr = ext_grid
x, y, z = np.mgrid[xl:xr:eval('{}j'.format(Nx)),
yl:yr:eval('{}j'.format(Ny)),
zl:zr:eval('{}j'.format(Nz))]
scalars = eval(expr)
src = mym.pipeline.scalar_field(x, y, z, scalars)
cont1 = mym.pipeline.iso_surface(src, color=col_isurf, contours=[0],
transparent=False, opacity=0.8)
cont1.compute_normals = False # for some reasons, setting this to true actually cause
# more unevenness on the surface, instead of more smooth
cont1.actor.property.specular = 0.2 #0.4 #0.8
cont1.actor.property.specular_power = 55.0 #15.0
def ensure_proj_z(plane_coeffs, min_z_proj):
a,b,c,d = plane_coeffs
if np.abs(c) < min_z_proj:
s = ((1 - min_z_proj**2) / (a**2 + b**2))**0.5
coeffs = np.array([s*a, s*b, np.sign(c)*min_z_proj, d])
assert np.abs(np.linalg.norm(coeffs[:3])-1) < 1e-3
return coeffs
return plane_coeffs
def isplanar(xyz,sample_neighbors,dist_thresh,num_inliers,z_proj):
"""
Checks if at-least FRAC_INLIERS fraction of points of XYZ (nx3)
points lie on a plane. The plane is fit using RANSAC.
XYZ : (nx3) array of 3D point coordinates
SAMPLE_NEIGHBORS : 5xN_RANSAC_TRIALS neighbourhood array
of indices into the XYZ array. i.e. the values in this
matrix range from 0 to number of points in XYZ
DIST_THRESH (default = 10cm): a point pt is an inlier iff dist(plane-pt)<dist_thresh
FRAC_INLIERS : fraction of total-points which should be inliers to
to declare that points are planar.
Z_PROJ : changes the surface normal, so that its projection on z axis is ATLEAST z_proj.
Returns:
None, if the data is not planar, else a 4-tuple of plane coeffs.
"""
frac_inliers = num_inliers/xyz.shape[0]
dv = -np.percentile(xyz,50,axis=0) # align the normal to face towards camera
max_iter = sample_neighbors.shape[-1]
plane_info = fit_plane_ransac(xyz,neighbors=sample_neighbors,
z_pos=dv,dist_inlier=dist_thresh,
min_inlier_frac=frac_inliers,nsample=20,
max_iter=max_iter)
if plane_info != None:
coeff, inliers = plane_info
coeff = ensure_proj_z(coeff, z_proj)
return coeff,inliers
else:
return #None
class DepthCamera(object):
"""
Camera functions for Depth-CNN camera.
"""
f = 520
@staticmethod
def plane2xyz(center, ij, plane):
"""
converts image pixel indices to xyz on the PLANE.
center : 2-tuple
ij : nx2 int array
plane : 4-tuple
return nx3 array.
"""
ij = np.atleast_2d(ij)
n = ij.shape[0]
ij = ij.astype('float')
xy_ray = (ij-center[None,:]) / DepthCamera.f
z = -plane[2]/(xy_ray.dot(plane[:2])+plane[3])
xyz = np.c_[xy_ray, np.ones(n)] * z[:,None]
return xyz
@staticmethod
def depth2xyz(depth):
"""
Convert a HxW depth image (float, in meters)
to XYZ (HxWx3).
y is along the height.
x is along the width.
"""
H,W = depth.shape
xx,yy = np.meshgrid(np.arange(W),np.arange(H))
X = (xx-W/2) * depth / DepthCamera.f
Y = (yy-H/2) * depth / DepthCamera.f
return np.dstack([X,Y,depth.copy()])
@staticmethod
def overlay(rgb,depth):
"""
overlay depth and rgb images for visualization:
"""
depth = depth - np.min(depth)
depth /= np.max(depth)
depth = (255*depth).astype('uint8')
return np.dstack([rgb[:,:,0],depth,rgb[:,:,1]])
def get_texture_score(img,masks,labels):
"""
gives a textureness-score
(low -> less texture, high -> more texture) for each mask.
"""
img = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
img = cv2.equalizeHist(img)
img = img.astype('float32')/255.0
img = sim.filters.gaussian_filter(img,sigma=1)
G = np.clip(np.abs(sim.filters.laplace(img)),0,1)
tex_score = []
for l in labels:
ts = np.sum(G[masks==l].flat)/np.sum((masks==l).flat)
tex_score.append(ts)
return np.array(tex_score)
def ssc(v):
"""
Returns the skew-symmetric cross-product matrix corresponding to v.
"""
v /= np.linalg.norm(v)
return np.array([[ 0, -v[2], v[1]],
[ v[2], 0, -v[0]],
[-v[1], v[0], 0]])
def rot3d(v1,v2):
"""
Rodrigues formula : find R_3x3 rotation matrix such that v2 = R*v1.
https://en.wikipedia.org/wiki/Rodrigues'_rotation_formula#Matrix_notation
"""
v1 /= np.linalg.norm(v1)
v2 /= np.linalg.norm(v2)
v3 = np.cross(v1,v2)
s = np.linalg.norm(v3)
c = v1.dot(v2)
Vx = ssc(v3)
return np.eye(3)+s*Vx+(1-c)*Vx.dot(Vx)
def unrotate2d(pts):
"""
PTS : nx3 array
finds principal axes of pts and gives a rotation matrix (2d)
to realign the axes of max variance to x,y.
"""
mu = np.median(pts,axis=0)
pts -= mu[None,:]
l,R = np.linalg.eig(pts.T.dot(pts))
R = R / np.linalg.norm(R,axis=0)[None,:]
# make R compatible with x-y axes:
if abs(R[0,0]) < abs(R[0,1]): #compare dot-products with [1,0].T
R = np.fliplr(R)
if not np.allclose(np.linalg.det(R),1):
if R[0,0]<0:
R[:,0] *= -1
elif R[1,1]<0:
R[:,1] *= -1
else:
print "Rotation matrix not understood"
return
if R[0,0]<0 and R[1,1]<0:
R *= -1
assert np.allclose(np.linalg.det(R),1)
# at this point "R" is a basis for the original (rotated) points.
# we need to return the inverse to "unrotate" the points:
return R.T #return the inverse