-
Notifications
You must be signed in to change notification settings - Fork 5
/
scoreAgent.py
323 lines (275 loc) · 12.5 KB
/
scoreAgent.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
import torch
import torch.nn as nn
from model import cornerModel, region_model
from drn import drn_c_26
from new_utils import *
import os
import torch.nn.functional as F
from config import config
class DoubleConv(nn.Module):
def __init__(self, in_channels, out_channels, batchnorm, mid_channels=None):
super(DoubleConv, self).__init__()
if not mid_channels:
mid_channels = out_channels
if batchnorm:
self.double_conv = nn.Sequential(
nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(mid_channels),
nn.LeakyReLU(inplace=True),
nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1),
nn.BatchNorm2d(out_channels),
nn.LeakyReLU(inplace=True))
else:
self.double_conv = nn.Sequential(
nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1),
nn.InstanceNorm2d(mid_channels),
nn.LeakyReLU(inplace=True),
nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1),
nn.InstanceNorm2d(out_channels),
nn.LeakyReLU(inplace=True))
def forward(self, x):
return self.double_conv(x)
class Down(nn.Module):
def __init__(self, in_channels, out_channels, batchnorm):
super(Down, self).__init__()
if batchnorm:
self.down_conv = nn.Sequential(
nn.Conv2d(in_channels, in_channels, kernel_size=3, padding=1, stride=2),
nn.BatchNorm2d(in_channels),
nn.LeakyReLU(inplace=True),
DoubleConv(in_channels, out_channels, batchnorm=batchnorm))
else:
self.down_conv = nn.Sequential(
nn.Conv2d(in_channels, in_channels, kernel_size=3, padding=1, stride=2),
nn.InstanceNorm2d(in_channels),
nn.LeakyReLU(inplace=True),
DoubleConv(in_channels, out_channels, batchnorm=batchnorm))
def forward(self, x):
return self.down_conv(x)
class Up(nn.Module):
def __init__(self, in_channels, out_channels, batchnorm, bilinear=False):
super(Up, self).__init__()
if bilinear:
self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
self.conv = DoubleConv(in_channels, out_channels, in_channels // 2, batchnorm=batchnorm)
else:
self.up = nn.ConvTranspose2d(in_channels, in_channels // 2, kernel_size=4, stride=2, padding=1)
self.conv = DoubleConv(in_channels, out_channels, batchnorm=batchnorm)
def forward(self, x1, x2):
x1 = self.up(x1)
x = torch.cat([x2, x1], dim=1)
return self.conv(x)
class UNet_big(nn.Module):
def __init__(self, n_channels, n_classes, batchnorm, useSigmoid=True, bilinear=False):
super(UNet_big, self).__init__()
self.n_channels = n_channels
self.n_classes = n_classes
self.bilinear = bilinear
self.inc = DoubleConv(n_channels, 32, batchnorm=batchnorm)
self.down1 = Down(32, 64, batchnorm=batchnorm)
self.down2 = Down(64, 128, batchnorm=batchnorm)
self.down3 = Down(128, 256, batchnorm=batchnorm)
self.down4 = Down(256, 512, batchnorm=batchnorm)
self.down5 = Down(512, 1024, batchnorm=batchnorm)
factor = 2 if bilinear else 1
self.down6 = Down(1024, 2048// factor, batchnorm=batchnorm)
self.up1 = Up(2048, 1024 // factor, batchnorm, bilinear)
self.up2 = Up(1024, 512 // factor, batchnorm, bilinear)
self.up3 = Up(512, 256 // factor, batchnorm, bilinear)
self.up4 = Up(256, 128 // factor, batchnorm, bilinear)
self.up5 = Up(128, 64 // factor, batchnorm, bilinear)
self.up6 = Up(64, 32, batchnorm, bilinear)
if useSigmoid:
self.out = nn.Sequential(
nn.Conv2d(32, n_classes, kernel_size=1),
nn.Sigmoid())
else:
self.out = nn.Sequential(
nn.Conv2d(32, n_classes, kernel_size=1),
nn.BatchNorm2d(n_classes),
nn.ReLU())
def forward(self, x):
x1 = self.inc(x)
x2 = self.down1(x1)
x3 = self.down2(x2)
x4 = self.down3(x3)
x5 = self.down4(x4)
x6 = self.down5(x5)
x7 = self.down6(x6)
x = self.up1(x7, x6)
x = self.up2(x, x5)
x = self.up3(x, x4)
x = self.up4(x, x3)
x = self.up5(x, x2)
x = self.up6(x, x1)
return self.out(x)
class UNnet(nn.Module):
def __init__(self, batchnorm, bilinear=False, backbone_channel=64):
super(UNnet, self).__init__()
self.bilinear = bilinear
self.backbone_channel = backbone_channel
self.inc = DoubleConv(2, 16, batchnorm=batchnorm)
self.down1 = Down(16+self.backbone_channel, 64, batchnorm=batchnorm)
self.down2 = Down(64, 128, batchnorm=batchnorm)
self.down3 = Down(128, 256, batchnorm=batchnorm)
self.down4 = Down(256, 512, batchnorm=batchnorm)
factor = 2 if bilinear else 1
self.down5 = Down(512, 1024 // factor, batchnorm=batchnorm)
self.up1 = Up(1024, 512 // factor, batchnorm, bilinear)
self.up2 = Up(512, 256 // factor, batchnorm, bilinear)
self.up3 = Up(256, 128 // factor, batchnorm, bilinear)
self.up4 = Up(128, 64 // factor, batchnorm, bilinear)
self.up51 = nn.ConvTranspose2d(64, 32, kernel_size=4, stride=2, padding=1)
self.up52 = DoubleConv(32+16, 32, batchnorm=batchnorm)
self.out = nn.Sequential(
nn.Conv2d(32, 1, kernel_size=1),
nn.Sigmoid())
def forward(self, mask, image_volume):
x1 = self.inc(mask)
x2 = self.down1(torch.cat((x1, image_volume),1))
x3 = self.down2(x2)
x4 = self.down3(x3)
x5 = self.down4(x4)
x6 = self.down5(x5)
x = self.up1(x6, x5)
x = self.up2(x, x4)
x = self.up3(x, x3)
x = self.up4(x, x2)
x = self.up51(x)
x = self.up52(torch.cat((x1,x),1))
return self.out(x)
class scoreEvaluator_with_train(nn.Module):
def __init__(self, datapath, device, backbone_channel=64, use_heat_map=True):
super(scoreEvaluator_with_train, self).__init__()
self.backbone_channel = backbone_channel
self.backbone = UNet_big(3, backbone_channel, batchnorm=True, useSigmoid=False) # for image
channel_size = backbone_channel# + bin_size if corner_bin else backbone_channel
channel_size = channel_size + 2# if use_heat_map else channel_size
self.cornerNet = UNnet(batchnorm=True, backbone_channel=channel_size)
self.edgeNet = UNnet(batchnorm=True, backbone_channel=channel_size)
self.img_cache = imgCache(datapath)
self.region_cache = regionCache(os.path.join(config['data_folder'], 'result/corner_edge_region/entire_region_mask'))
self.heatmapNet = UNet_big(3, 2, batchnorm=True, useSigmoid=True)
self.device = device
def getheatmap(self, img):
heatmap = self.heatmapNet(img)
return heatmap
def imgvolume(self, img):
imgvolume = self.backbone(img)
return imgvolume
def cornerEvaluator(self, mask, img_volume, heatmap):
'''
:param mask: graph mask Nx2xhxw
:return: Nx1xhxw
'''
volume = img_volume
volume = torch.cat((volume, heatmap), 1)
out = self.cornerNet(mask, volume)
return out
def edgeEvaluator(self, mask, img_volume, heatmap):
volume = img_volume
volume = torch.cat((volume, heatmap), 1)
out = self.edgeNet(mask, volume)
return out
def regionEvaluator(self):
pass
def corner_map2score(self, corners, corner_map):
corner_state = np.ones(corners.shape[0])
scale_corners = corners
for corner_i in range(scale_corners.shape[0]):
loc = np.round(scale_corners[corner_i]).astype(np.int)
if loc[0] <= 1:
x0 = 0
else:
x0 = loc[0] - 1
if loc[0] >= 254:
x1 = 256
else:
x1 = loc[0] + 2
if loc[1] <= 1:
y0 = 0
else:
y0 = loc[1] - 1
if loc[1] >= 254:
y1 = 256
else:
y1 = loc[1] + 2
heat = corner_map[x0:x1, y0:y1]
corner_state[corner_i] = 1-2*heat.sum()/(heat.shape[0]*heat.shape[1])
return corner_state
def get_score_list(self, candidate_list):
for candidate_ in candidate_list:
self.get_score(candidate_)
def get_score(self, candidate, img_volume=None, heatmap=None):
graph = candidate.graph
corners = graph.getCornersArray()
edges = graph.getEdgesArray()
if img_volume is None or heatmap is None:
img = self.img_cache.get_image(candidate.name)
img = img.transpose((2,0,1))
img = (img - np.array(config['mean'])[:, np.newaxis, np.newaxis]) / np.array(config['std'])[:, np.newaxis, np.newaxis]
img = torch.cuda.FloatTensor(img, device=self.device).unsqueeze(0)
# corner and image volume
with torch.no_grad():
img_volume = self.imgvolume(img)
heatmap = self.getheatmap(img)
mask = render(corners, edges, render_pad=-1, scale=1)
mask = torch.cuda.FloatTensor(mask, device=self.device).unsqueeze(0)
# corner and image volume
with torch.no_grad():
corner_pred = self.cornerEvaluator(mask, img_volume, heatmap)
corner_map = corner_pred.squeeze().cpu().detach().numpy()
corner_map = np.clip(corner_map, 0, 1)
corner_state = self.corner_map2score(corners, corner_map)
# corner score
graph.store_score(corner_score=corner_state)
gt_mask = mask.squeeze()[0].detach().cpu().numpy()>0
with torch.no_grad():
edge_pred = self.edgeEvaluator(mask, img_volume, heatmap)
pred_bad_edges = edge_pred.squeeze().detach().cpu().numpy()
pred_bad_edges = np.clip(pred_bad_edges, 0, 1)*gt_mask
for edge_ele in graph.getEdges():
loc1 = edge_ele.x[0].x
loc2 = edge_ele.x[1].x
loc1 = (round(loc1[0]), round(loc1[1]))
loc2 = (round(loc2[0]), round(loc2[1]))
edge_mask = cv2.line(
np.ones((int(256),int(256)))*0,
loc1[::-1], loc2[::-1], 1.0,
thickness=2)
ratio = np.sum(np.multiply(pred_bad_edges, edge_mask))/np.sum(edge_mask)
edge_ele.store_score(1-2*ratio)
# region
gt_mask = self.region_cache.get_region(candidate.name)
gt_mask = gt_mask > 0.4
conv_mask = render(corners=corners, edges=edges, render_pad=0, edge_linewidth=1)[0]
conv_mask = 1 - conv_mask
conv_mask = conv_mask.astype(np.uint8)
labels, region_mask = cv2.connectedComponents(conv_mask, connectivity=4)
background_label = region_mask[0,0]
all_masks = []
for region_i in range(1, labels):
if region_i == background_label:
continue
the_region = region_mask == region_i
if the_region.sum() < 20:
continue
all_masks.append(the_region)
pred_mask = (np.sum(all_masks, 0) + (1 - conv_mask))>0
iou = IOU(pred_mask, gt_mask)
region_score = np.array([iou])
graph.store_score(region_score=region_score)
def store_weight(self, path, prefix):
with open(os.path.join(path, '{}_{}.pt'.format(prefix, 'backbone')), 'wb') as f:
torch.save(self.backbone.state_dict(), f)
with open(os.path.join(path, '{}_{}.pt'.format(prefix, 'edge')), 'wb') as f:
torch.save(self.edgeNet.state_dict(), f)
with open(os.path.join(path, '{}_{}.pt'.format(prefix, 'corner')), 'wb') as f:
torch.save(self.cornerNet.state_dict(), f)
def load_weight(self, path, prefix):
with open(os.path.join(path, '{}_{}.pt'.format(prefix, 'backbone')), 'rb') as f:
self.backbone.load_state_dict(torch.load(f))
with open(os.path.join(path, '{}_{}.pt'.format(prefix, 'corner')), 'rb') as f:
self.cornerNet.load_state_dict(torch.load(f))
with open(os.path.join(path, '{}_{}.pt'.format(prefix, 'edge')), 'rb') as f:
self.edgeNet.load_state_dict(torch.load(f))