forked from mathieuorhan/pointnet2_semantic
-
Notifications
You must be signed in to change notification settings - Fork 0
/
visu.py
executable file
·267 lines (242 loc) · 12.8 KB
/
visu.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
"""
This file has two main functionalities : statistics (stats) and batch visualisation
Stats (call python visu.py --stats=true) will first give you statistics on the
class repartition of points (i.e. 10% for class 1, etc.) in the clouds resulting from
preprocessing. Then it produces some inputs from those clouds according to the dataset accessor
you are using, and computes their own class repartition, which is different from before because
some points are favored over others: sometimes because we want them to, and sometimes because we
can't help it. (In fact, the inputs that are generated are usually undersampled to fit into the
nb of points the network can process, but here we skip that step because it's a uniform sampling).
Then those inputs that were generated are used to give you other information : per point probability
of being taken as input when the point's scene is considered. A histogram of these probabilities is
saved and be plotted unless using the option --draw=False. Then all these points are exported with
their probabilities as one point cloud per scene for visualisation in CloudCompare for example. The seeds
(points around which a box is computed and the final input points are sampled in that box) are also
provided in a separate point cloud and may be visualised in CloudCompare for example by giving them
a large radius. In all, stats is a powerful tools that allows you to check that the sampling you're using
fully explores your point clouds.
Batch visualisation will draw batches exactly as they are fed into the network and it will align them on
one horizontal grid. This way you can quickly check whether the input of the network is distinguishable
by humans. You can set the number of batches you want with the --n option (set it to zero if you are not
interested by this functionalities).
This file is of course compatible with both semantic.py and scannet.py (or very nearly, and only
small modifications shound be necessary).
Since scannet.py implements the way the scannet dataset was fed into the network by the creators
of pointnet2, to excellent results, it is a good idea to compare with it.
"""
import argparse
import numpy as np
import os
import importlib
import json
import utils.pc_util as pc_util
# Parser
parser = argparse.ArgumentParser()
parser.add_argument('--set', default="train", help='train or test [default: train]')
parser.add_argument('--stats', type=bool, default=False, help='Stats visualisation on inputs [default: False]')
parser.add_argument('--draw', type=bool, default=True, help='Plots in stats visualisation [default: True]')
parser.add_argument('--n', type=int, default=8, help='Number of batches you want to visualise [default : 1]')
parser.add_argument('--nps', type=int, default=100, help='Number of inputs per scene when doing stats [default : 100]')
parser.add_argument('--batch_size', type=int, default=8, help='Batch Size [default: 32]')
parser.add_argument('--num_point', type=int, default=4096, help='Point Number [default: 8192]')
parser.add_argument('--dataset', default='semantic', help='Dataset [default: semantic_color]')
parser.add_argument('--config', type=str, default="config.json", metavar='N', help='config file')
FLAGS = parser.parse_args()
JSON_DATA_CUSTOM = open(FLAGS.config).read()
CUSTOM = json.loads(JSON_DATA_CUSTOM)
JSON_DATA = open('default.json').read()
PARAMS = json.loads(JSON_DATA)
PARAMS.update(CUSTOM)
SET = FLAGS.set
NBATCH = FLAGS.n
BATCH_SIZE = FLAGS.batch_size
NUM_POINT = FLAGS.num_point
DATASET_NAME = FLAGS.dataset
STATS = FLAGS.stats
DRAW_PLOTS = FLAGS.draw
STAT_INPUT_NB = FLAGS.nps # per scene of the dataset
DROPOUT = False
DROPOUT_RATIO = 0.875
AUGMENT = False
MAX_EXPORT = 20 # the maximum number of scenes to be exported
GROUP_BY_BATCHES = True
if DROPOUT:
print("dropout is on, with ratio %f" %(DROPOUT_RATIO))
if AUGMENT:
print ("rotation is on")
# Import dataset
data = importlib.import_module('dataset.' + DATASET_NAME)
DATA = data.Dataset(npoints=NUM_POINT, split=SET, box_size=PARAMS['box_size'], use_color=PARAMS['use_color'],
dropout_max=PARAMS['input_dropout'], path=PARAMS['data_path']
, z_feature=PARAMS['use_z_feature'])
NUM_CLASSES = DATA.num_classes
# Outputs
OUTPUT_DIR = os.path.join("visu", DATASET_NAME+"_"+SET)
if not os.path.exists("visu"): os.mkdir("visu")
if not os.path.exists(OUTPUT_DIR): os.mkdir(OUTPUT_DIR)
OUTPUT_DIR_HIST = os.path.join(OUTPUT_DIR, "hist")
OUTPUT_DIR_STATS = os.path.join(OUTPUT_DIR, "color_proba_selection")
OUTPUT_DIR_SEEDS = os.path.join(OUTPUT_DIR, "seeds")
OUTPUT_DIR_BATCHES = os.path.join(OUTPUT_DIR, "grouped_by_batches")
if STATS:
"""
Here we want to get info about class representation in decimated data
and in the data we feed to the network.
We also want to know whether the data is well fed by the sampling algorithm
because we don't want to focus too much on always the same points. Thus we
compute an estimator of the probability of being selected when a scene is
considered for input points, and we export the map of probability
(one per scene).
Moreover we export with each point cloud a cloud containing only the seeds. You
may visualise with CloudCompare by loading it separately from the cloud and
setting a big point size for the seeds.
"""
if not os.path.exists(OUTPUT_DIR_HIST): os.mkdir(OUTPUT_DIR_HIST)
if not os.path.exists(OUTPUT_DIR_STATS): os.mkdir(OUTPUT_DIR_STATS)
if not os.path.exists(OUTPUT_DIR_SEEDS): os.mkdir(OUTPUT_DIR_SEEDS)
import matplotlib.pyplot as plt
import matplotlib
# the histogram of the decimated point cloud labels
if DRAW_PLOTS:
font = {'family' : 'normal',
'weight' : 'bold',
'size' : 22}
matplotlib.rc('font', **font)
fig, ax = plt.subplots()
ax.set_xlim(-0.5, NUM_CLASSES + 0.5)
plt.xticks(range(NUM_CLASSES))
xtickNames = plt.setp(ax, xticklabels=DATA.short_labels_names)
plt.setp(xtickNames, rotation=25, fontsize=16)
label_hist = DATA.get_hist()
label_hist = np.array(label_hist)
density_hist = label_hist/np.sum(label_hist)
plt.bar(range(NUM_CLASSES), density_hist, color='green')
#plt.yscale('log')
#plt.xlabel('Classes : {}'.format(DATA.get_list_classes_str()))
plt.ylabel('Proportion')
#plt.title('Histogram of class repartition in decimated point clouds of set {}'.format(SET))
plt.draw()
plt.savefig(os.path.join(OUTPUT_DIR_HIST,"dec.png"))
# generate real inputs as fed into the network
filenames = DATA.get_data_filenames()
print("Generating {} inputs".format(STAT_INPUT_NB*len(filenames)))
pc_shapes=list()
hist_list = list() # counts for every point of every scene the nb of times it is taken as input
seeds_list=list()
label_hist = np.zeros(NUM_CLASSES)
scene_counters = np.zeros(len(filenames))
for f in range(len(filenames)):
pc_shapes.append(DATA.get_scene_shape(f))
hist_list.append(np.zeros(pc_shapes[f][0]))
seeds_list.append(list())
#if len(np.unique(np.array(pc_shapes))) < len(pc_shapes):
#print("WARNING : scenes cannot be distinguished by shape; data generated into "+OUTPUT_DIR_STATS+" and "+OUTPUT_DIR_SEEDS+" may be erroneous")
for i in range(STAT_INPUT_NB*len(filenames)):
if i%100==0 and i>0:
print("{} inputs generated".format(i))
f, input_mask, input_label_hist, seed_idx = DATA.next_input(DROPOUT, True, False, True)
hist_list[f]+=input_mask
scene_counters[f]+=1
seeds_list[f].append(seed_idx)
label_hist+=input_label_hist
# the histogram of the input point cloud labels
label_hist = np.array(label_hist)
density_hist = label_hist/np.sum(label_hist)
#fig, ax = plt.figure(2)
fig, ax = plt.subplots()
ax.set_xlim(-0.5, NUM_CLASSES + 0.5)
plt.xticks(range(NUM_CLASSES))
xtickNames = plt.setp(ax, xticklabels=DATA.short_labels_names)
plt.setp(xtickNames, rotation=25, fontsize=16)
plt.bar(range(NUM_CLASSES), density_hist, color='green')
plt.ylabel('Proportion')
#plt.title('Histogram of class repartition in input point clouds of set {}'.format(SET))
plt.draw()
plt.savefig(os.path.join(OUTPUT_DIR_HIST,"input.png"))
# the histogram of the probability of being fed into the network
filenamesForExport = filenames[0:min(len(filenames), MAX_EXPORT)]
for f, filename in enumerate(filenamesForExport):
if scene_counters[f]==0:
continue
if len(filenames) < 4:
plt.figure(3+f)
else:
plt.figure(3)
density = 100*hist_list[f]/scene_counters[f]
plt.hist(density, bins=20, color='green')
plt.yscale('log')
plt.xlabel(r'Probability of occurence in %')
plt.ylabel('Proportion')
plt.title('Histogram of selection likelihood in scene {} of set {}'.format(filename, SET))
plt.draw()
plt.savefig(os.path.join(OUTPUT_DIR_HIST,"proba_scene_"+str(f)+".png"))
if DRAW_PLOTS:
plt.show()
# exporting the probability of being selected and the point seeds.
print("exporting {} point clouds with densities".format(len(filenamesForExport)))
for f, filename in enumerate(filenamesForExport):
if np.sum(hist_list[f])==0:
continue
density = hist_list[f]/np.sum(hist_list[f])
point_cloud = DATA.get_scene(f)
# all points
np.savetxt(os.path.join(OUTPUT_DIR_STATS,"{}_{}_{}.txt".format(SET, os.path.basename(filename), "appearance_density")), np.hstack((point_cloud, density.reshape((-1,1)))), delimiter=" ")
# this time with a mask removing points that weren't seen at all
mask = density > 0
np.savetxt(os.path.join(OUTPUT_DIR_STATS,"{}_{}_{}.txt".format(SET, os.path.basename(filename), "appearance_density_cleaned")), np.hstack((point_cloud, density.reshape((-1,1))))[mask], delimiter=" ")
# now the seeds
np.savetxt(os.path.join(OUTPUT_DIR_SEEDS,"{}_{}_{}.txt".format(SET, os.path.basename(filename), "seeds")), np.array(seeds_list[f]), delimiter=" ")
if NBATCH > 0:
if not os.path.exists(OUTPUT_DIR_BATCHES): os.mkdir(OUTPUT_DIR_BATCHES)
print("batch visualisation :")
if GROUP_BY_BATCHES and NBATCH > 0:
"""
Here we export batches as they are fed into the network
The idea is to position the scenes of the different batches on a grid
where batches align themselves on the same x coordinates
"""
data, _, _ = DATA.next_batch(BATCH_SIZE, AUGMENT, DROPOUT)
data = data[0,:,0:3].reshape((-1,3))
# get spatial dimension of input
xmin, ymin, _ = np.min(data, axis=0)
xmax, ymax, _ = np.max(data, axis=0)
xsize = xmax - xmin + 15
ysize = ymax - ymin + 15
# initialize cloud and labels
if PARAMS['use_color']:
visu_cloud = np.array([0,0,0,0,0,0])
else:
visu_cloud = np.array([0,0,0])
visu_labels = np.array([0])
for i in range(NBATCH):
data, label_data, _ = DATA.next_batch(BATCH_SIZE, AUGMENT, DROPOUT)
print ("Processing batch number " + str(i))
if GROUP_BY_BATCHES:
positioned_data = list()
for j, scene in enumerate(data):
x = j*xsize
y = i * ysize
z = 0
if PARAMS['use_color']:
positioned_data.append(scene + np.array([x,y,z,0,0,0]) - np.min(scene, axis=0)*np.array([1,1,1,0,0,0]))
else:
positioned_data.append(scene + np.array([x,y,z]) - np.min(scene, axis=0))
batch_points = np.vstack(positioned_data)
visu_cloud = np.vstack((visu_cloud, batch_points))
batch_labels = np.hstack(label_data)
visu_labels = np.hstack((visu_labels, batch_labels))
else:
for j, scene in enumerate(data):
labels = label_data[j]
if PARAMS['use_color']:
pc_util.write_ply_true_color(scene[:,0:3], (255*scene[:,3:6]).astype(int), OUTPUT_DIR_BATCHES+"/{}_{}_{}.txt".format(SET, "trueColors", j))
else:
pc_util.write_ply_true_color(scene[:,0:3], np.zeros(scene[:,0:3].shape), OUTPUT_DIR_BATCHES+"/{}_{}_{}.txt".format(SET, "trueColors", j))
pc_util.write_ply_color(scene[:,0:3], labels, OUTPUT_DIR_BATCHES+"/{}_{}_{}.txt".format(SET, "labelColors", j), NUM_CLASSES)
if GROUP_BY_BATCHES and NBATCH > 0:
if PARAMS['use_color']:
pc_util.write_ply_true_color(visu_cloud[:,0:3], (255*visu_cloud[:,3:6]).astype(int), OUTPUT_DIR_BATCHES+"/{}_{}.txt".format(SET, "trueColors"))
else:
pc_util.write_ply_true_color(visu_cloud[:,0:3], np.zeros(visu_cloud[:,0:3].shape), OUTPUT_DIR_BATCHES+"/{}_{}.txt".format(SET, "trueColors"))
pc_util.write_ply_color(visu_cloud[:,0:3], visu_labels, OUTPUT_DIR_BATCHES+"/{}_{}.txt".format(SET, "labelColors"), NUM_CLASSES)
print("done")