-
Notifications
You must be signed in to change notification settings - Fork 37
/
utils_faster_rcnn.py
271 lines (227 loc) · 11 KB
/
utils_faster_rcnn.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
# Copyright (c) 2019 Uber Technologies, Inc.
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.
import skimage
import skimage.io
import skimage.transform
import numpy as np
from matplotlib.pyplot import *
from IPython import embed
from general.conv_calc import receptive_calc_conv, receptive_calc_conv_same
def plot_boxes(im, pos, box_coords=None, obj_scores=None):
assert len(im.shape) == 3, 'only supports one img plotting'
gray()
subplot(111)
imshow(im.squeeze())
# show true box
for i in range(pos.shape[0]):
i_center, j_center, height, width = pos[i]
gca().add_patch(Rectangle((j_center - width/2, i_center - height/2), width, height, ec='m', lw=2.5, fc=(0,0,0,0)))
# show output boxes
if box_coords is not None:
for i in range(box_coords.shape[0]):
bi_c, bj_c, bh, bw = box_coords[i]
gca().add_patch(Rectangle((bj_c - bw/2, bi_c - bh/2), bw, bh, ec='g', fc=(0,0,0,0)))
if obj_scores is not None:
text(bj_c - bw/2, bi_c - bh/2, '%.2f,%.2f'%(obj_scores[i][0],obj_scores[i][1]), color='green')
show()
return
def plot_pos_boxes(im, pos, box_coords, pos_scores, showlabel=True):
assert len(im.shape) == 3, 'only supports one img plotting'
gray()
#subplot(111)
imshow(im.squeeze())
# show true box
for i in range(pos.shape[0]):
i_center, j_center, height, width = pos[i]
gca().add_patch(Rectangle((j_center - width/2, i_center - height/2), width, height, ec='m', lw=2.5, fc=(0,0,0,0)))
# show output boxes
for i in range(box_coords.shape[0]):
bi_c, bj_c, bh, bw = box_coords[i]
gca().add_patch(Rectangle((bj_c - bw/2, bi_c - bh/2), bw, bh, ec='g', fc=(0,0,0,0)))
if showlabel:
text(bj_c - bw/2, bi_c - bh/2, '%d'%(pos_scores[i]), color='green')
else:
text(bj_c - bw/2, bi_c - bh/2, '%.2f'%(pos_scores[i]), color='green')
#show()
return
def plot_boxes_pos_neg(im, gt, pos_box, neg_box):
assert len(im.shape) == 3, 'only supports one img plotting'
gray()
#subplot(111)
imshow(im.squeeze())
# show true box
for i in range(gt.shape[0]):
i_center, j_center, height, width = gt[i]
gca().add_patch(Rectangle((j_center - width/2, i_center - height/2), width, height, ec='m', lw=2.5, fc=(0,0,0,0)))
# show output boxes
for i in range(pos_box.shape[0]):
bi_c, bj_c, bh, bw = pos_box[i]
gca().add_patch(Rectangle((bj_c - bw/2, bi_c - bh/2), bw, bh, ec='g', lw=1.5, fc=(0,0,0,0)))
for j in range(neg_box.shape[0]):
bi_c, bj_c, bh, bw = neg_box[j]
gca().add_patch(Rectangle((bj_c - bw/2, bi_c - bh/2), bw, bh, ec='r', fc=(0,0,0,0)))
#show()
return
def plot_pos_boxes_thickness(im, pos, box_coords, pos_scores):
assert len(im.shape) == 3, 'only supports one img plotting'
gray()
#subplot(111)
imshow(im.squeeze())
# show true box
for i in range(pos.shape[0]):
i_center, j_center, height, width = pos[i]
gca().add_patch(Rectangle((j_center - width/2, i_center - height/2), width, height, ec='m', lw=2.5, fc=(0,0,0,0)))
# show output boxes
for i in range(box_coords.shape[0]):
bi_c, bj_c, bh, bw = box_coords[i]
gca().add_patch(Rectangle((bj_c - bw/2, bi_c - bh/2), bw, bh, ec='g', lw=pos_scores[i], fc=(0,0,0,0)))
#if showlabel:
# text(bj_c - bw/2, bi_c - bh/2, '%d'%(pos_scores[i]), color='green')
#else:
# text(bj_c - bw/2, bi_c - bh/2, '%.2f'%(pos_scores[i]), color='green')
#show()
return
def make_anchors_numpy(featuremap_hw, batch_size, anchor_templates, receptsize_stride_offset):
# step 1: compute field center
# derived from net_utils.lua in Densecap code
# calculates how to map each point on feature map back to original img
#
# For simple conv(5).pool(2).conv(5).pool(2) model: ( 16, 4, 0)
# For VGG without final pool layer: (196, 16, -90)
recept_size, stride, offset = receptsize_stride_offset
i0 = j0 = recept_size/2
si = sj = stride
# step 2: make anchors in terms of coords on original img, based on current featuremap locations and anchor templates
# anchor_templates: (k,2) array presenting k anchor box templates in (box height, box width)
# return anchors: (B, 4k, h, w) numpy array (dtype=float32), if reshape to (B, k, 4, h, w) then along the 3rd dim we have (ic, jc, bh, bw)
h, w = featuremap_hw
i_centers = np.arange(0, float(h)) # i_centers in the feature map
i_centers *= si
i_centers += i0 + offset # now tranformed to i_centers in original image
j_centers = np.arange(0, float(w))
j_centers *= sj
j_centers += j0 + offset
box_heights = anchor_templates[:,0]
box_widths = anchor_templates[:,1]
print('ic',i_centers)
print('jc',j_centers)
print('bh',box_heights)
print('bw',box_widths)
# empty output
k = len(anchor_templates)
anchors = np.empty([4, k, h, w])
anchors[0] = np.tile(i_centers.reshape((1,-1,1)), (k,1,w)) # ic
anchors[1] = np.tile(j_centers.reshape((1,1,-1)), (k,h,1)) # jc
anchors[2] = np.tile(box_heights.reshape((-1,1,1)), (1,h,w)) # box_h
anchors[3] = np.tile(box_widths.reshape((-1,1,1)), (1,h,w)) # box_w
anchors = np.expand_dims(anchors,axis=0)
anchors = np.tile(anchors,(batch_size,1,1,1,1)) # (B, 4, k, h, w)
anchors = np.reshape(anchors, (batch_size, 4*k, h, w)) # (B, 4*k, h, w)
anchors = np.transpose(anchors, (0, 2, 3, 1)) # (B, h, w, 4*k)
return anchors.astype('float32')
def dense_to_one_hot(labels_dense, num_class=10):
# for more than one digits in a image
b, n = np.shape(labels_dense)
index_offset = np.arange(b*n) * num_class
labels_one_hot = np.zeros((b, n, num_class))
labels_one_hot.flat[index_offset + labels_dense.ravel()] = 1
return labels_one_hot
def make_anchors_mnist(featuremap_hw, batch_size, anchor_templates):
def calc_conv(w,s):
return lambda rs: (rs[0]+rs[1]*(w-1), rs[1]*s)
def calc_pool(w,s):
return lambda rs: (rs[0]+rs[1]*(w-1), rs[1]*s)
# step 1: compute field center
# derived from net_utils.lua in Densecap code
# calculates how to map each point on feature map back to original img
# hardcoded (for now) each layer's filter size and stride information
#
rs = (1.,1.)
rs = calc_conv(5,1)(rs) # 1st conv
rs = calc_pool(2,2)(rs) # 1st pool
rs = calc_conv(5,1)(rs) # 2nd conv
rs = calc_pool(2,2)(rs) # 2nd pool
i0 = j0 = rs[0]/2
si = sj = rs[1]
# step 2: make anchors in terms of coords on original img, based on current featuremap locations and anchor templates
# anchor_templates: (k,2) array presenting k anchor box templates in (box height, box width)
# return anchors: (B, 4k, h, w) numpy array (dtype=float32), if reshape to (B, k, 4, h, w) then along the 3rd dim we have (ic, jc, bh, bw)
h, w = featuremap_hw
i_centers = np.arange(0, float(h)) # i_centers in the feature map
i_centers *= si
i_centers += i0 # now tranformed to i_centers in original image
j_centers = np.arange(0, float(w))
j_centers *= sj
j_centers += j0
box_heights = anchor_templates[:,0]
box_widths = anchor_templates[:,1]
print('ic',i_centers)
print('jc',j_centers)
print('bh',box_heights)
print('bw',box_widths)
# empty output
k = len(anchor_templates)
anchors = np.empty([4, k, h, w])
anchors[0] = np.tile(i_centers.reshape((1,-1,1)), (k,1,w)) # ic
anchors[1] = np.tile(j_centers.reshape((1,1,-1)), (k,h,1)) # jc
anchors[2] = np.tile(box_heights.reshape((-1,1,1)), (1,h,w)) # box_h
anchors[3] = np.tile(box_widths.reshape((-1,1,1)), (1,h,w)) # box_w
anchors = np.expand_dims(anchors,axis=0)
anchors = np.tile(anchors,(batch_size,1,1,1,1)) # (B, 4, k, h, w)
anchors = np.reshape(anchors, (batch_size, 4*k, h, w)) # (B, 4*k, h, w)
anchors = np.transpose(anchors, (0, 2, 3, 1)) # (B, h, w, 4*k)
return anchors.astype('float32')
def make_anchors_mnist_same(featuremap_hw, batch_size, anchor_templates):
rso = (1.,1., 0.)
rso = receptive_calc_conv_same(5,1)(rso)
rso = receptive_calc_conv(2,2)(rso) # 1st pool
rso = receptive_calc_conv_same(5,1)(rso) # 2nd conv
rso = receptive_calc_conv(2,2)(rso) # 2nd pool
recept_size, stride, offset = rso
i0 = j0 = recept_size/2
si = sj = stride
# step 2: make anchors in terms of coords on original img, based on current featuremap locations and anchor templates
# anchor_templates: (k,2) array presenting k anchor box templates in (box height, box width)
# return anchors: (B, 4k, h, w) numpy array (dtype=float32), if reshape to (B, k, 4, h, w) then along the 3rd dim we have (ic, jc, bh, bw)
h, w = featuremap_hw
i_centers = np.arange(0, float(h)) # i_centers in the feature map
i_centers *= si
i_centers += i0 + offset # now tranformed to i_centers in original image
j_centers = np.arange(0, float(w))
j_centers *= sj
j_centers += j0 + offset
box_heights = anchor_templates[:,0]
box_widths = anchor_templates[:,1]
print('ic',i_centers)
print('jc',j_centers)
print('bh',box_heights)
print('bw',box_widths)
# empty output
k = len(anchor_templates)
anchors = np.empty([4, k, h, w])
anchors[0] = np.tile(i_centers.reshape((1,-1,1)), (k,1,w)) # ic
anchors[1] = np.tile(j_centers.reshape((1,1,-1)), (k,h,1)) # jc
anchors[2] = np.tile(box_heights.reshape((-1,1,1)), (1,h,w)) # box_h
anchors[3] = np.tile(box_widths.reshape((-1,1,1)), (1,h,w)) # box_w
anchors = np.expand_dims(anchors,axis=0)
anchors = np.tile(anchors,(batch_size,1,1,1,1)) # (B, 4, k, h, w)
anchors = np.reshape(anchors, (batch_size, 4*k, h, w)) # (B, 4*k, h, w)
anchors = np.transpose(anchors, (0, 2, 3, 1)) # (B, h, w, 4*k)
return anchors.astype('float32')