forked from siyandong/NeuralRouting
-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
249 lines (231 loc) · 10.3 KB
/
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
import numpy as np
import random, pickle, torch, copy, cv2
from glob import glob
from sklearn.cluster import MeanShift, estimate_bandwidth
from sklearn import mixture
from tqdm import tqdm
import scene_partition_tree
import dataset_loader
from config import *
# downsample.
g_n_pts_per_frame = opt.g_n_pts_per_frame
sampled_image_width, sampled_image_height = int(opt.image_width / 4), int(opt.image_height / 4)
# output directory.
g_prefix_predictionsImage = 'gmm_prediction/_{}_{}lvs_top{}_{}_step{}'.format(
opt.exp_name, opt.tree_height-1, opt.multi_leaf, opt.test_seq, opt.idx_step).replace(' ', '')
g_prefix_predictionsImage = g_prefix_predictionsImage + '_{}pts'.format(g_n_pts_per_frame)
if not os.path.exists(g_prefix_predictionsImage):
os.makedirs(g_prefix_predictionsImage)
# the data to evaluate.
crt_data_root = os.path.join(opt.data_root, opt.test_seq)
crt_frame_num = len(glob(os.path.join(crt_data_root, opt.color_format.format(0).replace('000000', '??????'))))
crt_frame_list = range(crt_frame_num)
def union_gmm(gmms, confs=None):
n = len(gmms)
if confs == None:
# confs = [1. / n] * n
confs = [1.] * n
gmm = scene_partition_tree.GMM(None,
np.concatenate([gmm.means for gmm in gmms], axis=0),
np.concatenate([gmm.covars for gmm in gmms], axis=0),
np.concatenate([(gmm.weights * conf)[:, np.newaxis]
for gmm, conf in zip(gmms, confs)]).flatten())
return gmm
def write_predictionsImage(path, predictions):
file = open(path, 'w')
count = 0 # num of valid prediction.
for idx in range(len(predictions)):
if predictions[idx] is None:
continue
count += 1
file.write('{}\n'.format(count))
for idx in range(len(predictions)):
if predictions[idx] is None:
continue
# gmm = predictions[idx]
gmm = predictions[idx][0]
confidence = predictions[idx][1]
file.write('{}\n{}\n{}\n'.format(idx, confidence, len(gmm.means))) # idx, confidence, n_cluster.
for cid in range(len(gmm.means)):
file.write('{} '.format(gmm.weights[cid])) # weight.
file.write('0 0 0 {} {} {}\n'.format(gmm.means[cid][0], gmm.means[cid][1], gmm.means[cid][2])) # w/o color, only position.
inv_corvar = np.linalg.inv(gmm.covars[cid]) # inv_corvar.
file.write('{} {} {} {} {} {} {} {} {}\n'.format(
inv_corvar[0, 0], inv_corvar[0, 1], inv_corvar[0, 2],
inv_corvar[1, 0], inv_corvar[1, 1], inv_corvar[1, 2],
inv_corvar[2, 0], inv_corvar[2, 1], inv_corvar[2, 2]))
file.close()
if __name__ == '__main__':
# build scene space partition tree structure.
nt = scene_partition_tree.NetTree(opt.scene_bbx, opt=opt)
nt.build_multi_level(opt.tree_height)
# compute gmm in tree leaf nodes.
if True:
# save raw coords in leaf nodes.
print('initializing leaf coords...')
if not g_bool_load_from_pickle:
nt.init_leaf_coords(train_data_root, train_frame_list)
nt.save_leaf_coords_to_file(g_pickle_leaf_coords_prefix)
else:
nt.load_leaf_coords_from_file(g_pickle_leaf_coords_prefix)
# fit gmm in each leaf node.
print('fitting leaf gmms...')
node_list = []
nt.get_node_list(nt.root_node, node_list)
valid_num = 0
for index, node in enumerate(tqdm(node_list)):
# if load gmm from file.
leaf_gmm_path = '{}/{}_leaf_gmm.pk'.format(g_pickle_leaf_gmm_prefix, node.node_id)
if os.path.exists(leaf_gmm_path):
with open(leaf_gmm_path, 'rb') as fp:
node.leaf_gmm = pickle.loads(fp.read())
valid_num+=1
continue
# fit gmm.
if True:
if len(node.leaf_coords) == 0: continue
for lid in range(len(node.leaf_coords)):
coords = node.leaf_coords[lid]
if len(coords) < 10:
node.leaf_gmm.append(None)
continue
crd_ary = np.array(coords)
# clustering as initialization.
bandwidth = estimate_bandwidth(crd_ary[:, 0:3], quantile=0.2, n_samples=500)
bandwidth *= 0.5
bandwidth = 0.05
ms = MeanShift(bandwidth=bandwidth, bin_seeding=True)
ms.fit(crd_ary[:, 0:3])
labels = ms.labels_
cluster_centers = ms.cluster_centers_
labels_unique = np.unique(labels)
n_cluster = len(labels_unique)
# fit gmm based on clusters.
gmm = mixture.GaussianMixture(n_components=n_cluster)
gmm.fit(crd_ary[:, 0:3])
means = gmm.means_
covars = gmm.covariances_
weights = gmm.weights_
# save to the leaf node.
node.leaf_gmm.append(scene_partition_tree.GMM(gmm, means, covars, weights))
# check.
if not len(node.leaf_coords) == len(node.leaf_gmm):
print('error: please clear gmm files and re-fit.')
exit()
# save to file.
with open(leaf_gmm_path, 'wb') as fp:
pickle.dump(node.leaf_gmm, fp)
#print('saved leaf gmm to file.')
print('{} valid pickle files.'.format(valid_num))
# set neural routing functions: load from trained checkpoint.
nt.initialize_levels()
print('routing functions loading checkpoint...')
for lid in range(opt.tree_height - 1):
file_prefix = '{}/l{}'.format(g_checkpoint_dir, lid)
file_suffix = 'step{}'.format(n_epoch[lid] * max_step_per_epoch)
nt.levels[lid].load_checkpoint(file_prefix, file_suffix)
# dataset.
the_list = crt_frame_list
the_list = the_list[::opt.idx_step]
dataset = dataset_loader.TestDataset_PPF(crt_data_root, the_list, g_n_pts_per_frame, neighbor_da2d=opt.n2d_lists)
loader = dataset_loader.DataLoader(
dataset=dataset,
batch_size=1,
shuffle=False,
num_workers=opt.num_workers)
# inference.
print('inferring correspondences...')
ps_remain, ps_correct = [], []
inlier_recalls, inlier_precisions, outlier_recalls = [], [], []
topk_inlier_recalls, topk_inlier_precisions, topk_outlier_recalls = [], [], []
p_rejections_wrong_inlier, p_rejections_outlier = [], []
n_metric_list = []
n_correct_list = []
for step, batch_sample in enumerate(tqdm(loader)):
route_labs = None
pt_in = batch_sample[0][0]
nb_ms_in = batch_sample[1][0]
if not type(batch_sample[2]) == type(-1):
route_labs = batch_sample[2][0].to(torch.int32)
fid = batch_sample[3][0].int().item()
rc_list = batch_sample[4][0]
batch_size = rc_list.shape[0]
predImg_path = '{}/predictionsImage_{}.txt'.format(g_prefix_predictionsImage, fid)
# predict route.
# to handle large batch size. 2021/03/12.
if batch_size <= 2048:
confidence, routes_pred = nt.inference(pt_in, nb_ms_in, beam=1, multi_leaf=opt.multi_leaf)
else:
confidence, routes_pred = nt.inference(
pt_in[0:2048],
nb_ms_in[0:2048],
beam=1, multi_leaf=opt.multi_leaf)
for sid_beg in range(2048, batch_size, 2048):
sid_end = sid_beg + 2048
if sid_end > batch_size: sid_end = batch_size
conf, rout = nt.inference(
pt_in[sid_beg:sid_end],
nb_ms_in[sid_beg:sid_end],
beam=1, multi_leaf=opt.multi_leaf)
confidence = np.concatenate((confidence, conf), axis=0)
routes_pred = np.concatenate((routes_pred, rout), axis=0)
# to handle large batch size. 2021/03/12.
beam_size = routes_pred.shape[1]
# save prediction.
predictions = []
for i in range(sampled_image_width * sampled_image_height):
predictions.append(None)
for sid in range(batch_size): # for each sample.
valids = [True] * beam_size
routes = []
for kid in range(beam_size): routes.append([])
for kid in range(beam_size):
for lid in range(opt.tree_height - 1):
if not valids[kid]:
break
if routes_pred[sid][kid][lid] == ary:
valids[kid] = False
continue
routes[kid].append(routes_pred[sid][kid][lid])
valid = False
for kid in range(routes_pred.shape[1]):
if valids[kid]:
valid = True
break
if valid:
# get the last level node.
gmms = []
confs = []
for kid in range(beam_size):
if valids[kid]:
node = nt.get_node(routes[kid][0:-1])
if len(node.leaf_gmm) == 0:
#print('no leaf.')
continue
gmm = node.leaf_gmm[routes[kid][-1]]
if gmm is None:
#print('no gmm.')
continue
if len(gmm.means) == 0:
#print('no mode.')
continue
if len(gmm.means) >= 20:
#print('invalid gmm.')
continue
gmms.append(gmm)
confs.append(confidence[sid][kid])
# merge gmm
if len(gmms) == 0:
continue
gmm = union_gmm(gmms, confs if 1 else None)
if gmm == None:
continue
# record
r, c = rc_list[sid].int()
idx = r * sampled_image_width + c
predictions[idx] = (gmm, 0.8)
# write to file
write_predictionsImage(predImg_path, predictions)
#print('frame {} saved.'.format(fid))
print('done.')
exit()