forked from T2Kim/DeepVirtualMarkers
/
Labeler.py
286 lines (237 loc) · 12.1 KB
/
Labeler.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
import numpy as np
from urllib.request import urlretrieve
import torch
import torch.nn.functional as F
import MinkowskiEngine as ME
try:
import open3d as o3d
except ImportError:
raise ImportError('Please install open3d with `pip install open3d`.')
import sys
sys.path.append("../")
sys.path.append("./")
from models import load_model
from models.res16unet import Res16UNet34C
# from lib.renderer import *
from lib.utils import precision_at_one
import json
import progressbar
import igl
import cv2
import os
import copy
from sklearn.neighbors import KDTree
from scipy import spatial
from matplotlib import cm
def pcd_sparse_tensor(coords, voxel_size=0.01):
# Create a batch, this process is done in a data loader during training in parallel.
batch = [pcd_voxel(coords, voxel_size)]
coordinates_, featrues_ = list(zip(*batch))
coordinates, features = ME.utils.sparse_collate(coordinates_, featrues_)
# Normalize features and create a sparse tensor
return coordinates, (features - 0.5).float()
def pcd_voxel(coords, voxel_size):
feats = np.zeros_like(coords, dtype=np.float32)
quantized_coords = np.floor(coords / voxel_size)
inds = ME.utils.sparse_quantize(quantized_coords, return_index=True)
return quantized_coords[inds], feats[inds]
class Labeler:
def __init__(self, config):
self.config = config
self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# Define a model and load the weights
NetClass = load_model(config.model)
self.model = NetClass(3, config.marker, config).to(self.device)
model_dict = torch.load(config.weights)
self.model.load_state_dict(model_dict['state_dict'])
self.model.eval()
self.voxel_size = config.voxel_size
self.marker = config.marker
self._init_color_map()
self.cmap_name = "default"
self.image_intrinsic = o3d.camera.PinholeCameraIntrinsic(o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault)
def getLabel(self, filename, is_mesh = True, y_filp = True, additional_aff = np.eye(3)):
if y_filp:
additional_aff = np.dot(additional_aff, np.array([[1.,0,0],[0,-1,0],[0,0,-1]]))
if is_mesh:
v, f = igl.read_triangle_mesh(filename)
mesh = o3d.geometry.TriangleMesh()
mesh.vertices = o3d.utility.Vector3dVector(v)
mesh.triangles = o3d.utility.Vector3iVector(f)
mesh.compute_vertex_normals()
tmp_pcd = mesh.sample_points_uniformly(100000)
sample_points = np.dot(np.asarray(tmp_pcd.points), additional_aff)
faces = f
points = v
else:
d_arr = cv2.imread(filename, cv2.IMREAD_UNCHANGED).astype(np.float32)
d_arr = cv2.bilateralFilter(d_arr, 10, 50, 1)
d_arr /= 1000.0
depth_raw = o3d.geometry.Image(d_arr)
pcd = o3d.geometry.PointCloud.create_from_depth_image(depth_raw, self.image_intrinsic)
if y_filp:
pcd.transform([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
pcd.estimate_normals(search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.05, max_nn=40))
pcd.orient_normals_towards_camera_location(camera_location=np.array([0., 0., 0.]))
sample_points = np.dot(np.asarray(pcd.points), additional_aff)
points = np.asarray(pcd.points)
pred, up_coord = self._inference(sample_points)
up_coord_colors = self._gen_soft_color(pred)
tar_m_pcd = o3d.geometry.PointCloud()
tar_m_pcd.points = o3d.utility.Vector3dVector(np.dot((up_coord * self.voxel_size), np.linalg.inv(additional_aff)))
tar_m_pcd_tree = o3d.geometry.KDTreeFlann(tar_m_pcd)
mesh_p_bag = {}
n_points = points.shape[0]
mesh_p_bag["pred"] = np.zeros((n_points, self.config.marker))
mesh_p_bag["color"] = np.zeros((n_points, 3))
mesh_p_bag["coord"] = points
if is_mesh:
mesh_p_bag["indices"] = faces
print("KNN Search")
bar = progressbar.ProgressBar(maxval=n_points,widgets=[' [', progressbar.Timer(), '] ', progressbar.Bar(), ' (', progressbar.ETA(), ') ',]).start()
for i, p in enumerate(points):
near_indices = tar_m_pcd_tree.search_hybrid_vector_3d(p, self.voxel_size * 3.4, 20)
if near_indices[0] < 1:
near_indices = tar_m_pcd_tree.search_knn_vector_3d(p, 1)
if (np.min(near_indices[2]) == 0):
min_idx = np.argmin(near_indices[2])
tmp_pred = pred[near_indices[1][min_idx], :]
tmp_color = up_coord_colors[near_indices[1][min_idx], :]
else:
dist_weights = np.reciprocal(near_indices[2])
dist_weights.reshape((1, -1))
tmp_pred = np.dot(dist_weights, pred[near_indices[1]]) / np.sum(dist_weights)
tmp_color = np.dot(dist_weights, up_coord_colors[near_indices[1]]) / np.sum(dist_weights)
mesh_p_bag["pred"][i, :] = tmp_pred.reshape((1, -1))
mesh_p_bag["color"][i, :] = tmp_color.reshape((1, -1))
bar.update(i)
bar.finish()
if is_mesh:
mesh.vertex_colors = o3d.utility.Vector3dVector(mesh_p_bag["color"])
mesh_p_bag["o3d_vim"] = [mesh]
else:
pcd.colors = o3d.utility.Vector3dVector(mesh_p_bag["color"])
mesh_p_bag["o3d_vim"] = [pcd]
# generate Virtual Marker sphere
pred_max_idx = np.argsort(mesh_p_bag["pred"], axis=0)[-1, :]
arr_w = list(range(mesh_p_bag["pred"].shape[1]))
pred_max_coords = mesh_p_bag["coord"][pred_max_idx, :]
pred_max_w = mesh_p_bag["pred"][pred_max_idx, arr_w]
for p, w in zip(pred_max_coords, pred_max_w):
if(w > 0.3):
mesh_sphere = o3d.geometry.TriangleMesh.create_sphere(radius=0.01)
mesh_sphere.compute_vertex_normals()
mesh_sphere.vertices = o3d.utility.Vector3dVector(np.asarray(mesh_sphere.vertices) + np.array(p))
mesh_sphere.paint_uniform_color([0.1, 0.1, 0.1])
mesh_p_bag["o3d_vim"].append(mesh_sphere)
else:
mesh_sphere = o3d.geometry.TriangleMesh()
mesh_p_bag["o3d_vim"].append(mesh_sphere)
return mesh_p_bag
def getLabelVoxel(self, filename, is_mesh = True, y_filp = True, additional_aff = np.eye(3)):
if y_filp:
additional_aff = np.dot(additional_aff, np.array([[1.,0,0],[0,-1,0],[0,0,-1]]))
if is_mesh:
v, f = igl.read_triangle_mesh(filename)
mesh = o3d.geometry.TriangleMesh()
mesh.vertices = o3d.utility.Vector3dVector(v)
mesh.triangles = o3d.utility.Vector3iVector(f)
mesh.compute_vertex_normals()
tmp_pcd = mesh.sample_points_uniformly(400000)
sample_points = np.dot(np.asarray(tmp_pcd.points), additional_aff)
points = np.asarray(mesh.vertices)
else:
d_arr = cv2.imread(filename, cv2.IMREAD_UNCHANGED).astype(np.float32)
d_arr = cv2.bilateralFilter(d_arr, 10, 50, 1)
d_arr /= 1000.0
depth_raw = o3d.geometry.Image(d_arr)
pcd = o3d.geometry.PointCloud.create_from_depth_image(depth_raw, self.image_intrinsic)
if y_filp:
pcd.transform([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
pcd.estimate_normals(search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.05, max_nn=40))
pcd.orient_normals_towards_camera_location(camera_location=np.array([0., 0., 0.]))
sample_points = np.dot(np.asarray(pcd.points), additional_aff)
points = np.asarray(pcd.points)
pred, up_coord = self._inference(sample_points)
tar_m_pcd = o3d.geometry.PointCloud()
points = np.dot((up_coord * self.voxel_size), np.linalg.inv(additional_aff))
tar_m_pcd.points = o3d.utility.Vector3dVector(points)
mesh_p_bag = {}
n_points = points.shape[0]
mesh_p_bag["pred"] = pred
mesh_p_bag["color"] = self._gen_soft_color(pred)
mesh_p_bag["coord"] = points
tar_m_pcd.colors = o3d.utility.Vector3dVector(mesh_p_bag["color"])
tar_m_pcd.estimate_normals(search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.05, max_nn=40))
mesh_p_bag["o3d_vim"] = [tar_m_pcd]
# generate Virtual Marker sphere
pred_max_idx = np.argsort(mesh_p_bag["pred"], axis=0)[-1, :]
arr_w = list(range(mesh_p_bag["pred"].shape[1]))
pred_max_coords = mesh_p_bag["coord"][pred_max_idx, :]
pred_max_w = mesh_p_bag["pred"][pred_max_idx, arr_w]
# pred_max_coords = pred_max_coords[np.where(pred_max_w > 0.3), :].reshape((-1,3))
for p, w in zip(pred_max_coords, pred_max_w):
if(w > 0.3):
mesh_sphere = o3d.geometry.TriangleMesh.create_sphere(radius=0.01)
mesh_sphere.compute_vertex_normals()
mesh_sphere.vertices = o3d.utility.Vector3dVector(np.asarray(mesh_sphere.vertices) + np.array(p))
mesh_sphere.paint_uniform_color([0.1, 0.1, 0.1])
mesh_p_bag["o3d_vim"].append(mesh_sphere)
else:
mesh_sphere = o3d.geometry.TriangleMesh()
mesh_p_bag["o3d_vim"].append(mesh_sphere)
return mesh_p_bag
def setIntrinsic(self, d_intrinsic):
self.image_intrinsic.set_intrinsics(d_intrinsic['width'],
d_intrinsic['height'],
d_intrinsic['fx'],
d_intrinsic['fy'],
d_intrinsic['cx'],
d_intrinsic['cy'])
def _inference(self, coords):
with torch.no_grad():
coordinates, features = pcd_sparse_tensor(coords, voxel_size=self.voxel_size)
sinput = ME.SparseTensor(features, coords=coordinates).to(self.device)
soutput = self.model(sinput)
pred = F.softmax(soutput.F[:, :self.marker], dim=1)
pred = pred.cpu().numpy()
cell = soutput.C.numpy()[:, 1:]
return pred, cell
def _gen_soft_color(self, upsampled_pred):
v_color = np.zeros((len(upsampled_pred), 3), dtype=np.float32)
for i in range(upsampled_pred.shape[1]):
v_color += np.dot((upsampled_pred[:, i])[:, np.newaxis],
np.array(self.COLOR_MAPS[self.cmap_name][i], dtype=np.float32).reshape(1, 3))
return v_color / 255
def _init_color_map(self):
self.COLOR_MAPS = {}
## default
cc_map = cm.get_cmap('Paired')
COLOR_MAP = np.zeros((100, 3))
part = [[],[],[],[],[],[]]
part[0]= np.array([12,6,7,8,9,10,11,0,1,2,3,4,5, 13,14,15,16])[::-1] # head
part[1]=np.concatenate((np.arange(17,25),np.arange(45,69))) # body
part[2]= np.arange(25,35) # right arm
part[3]= np.arange(35,45) # left arm
part[4]= np.arange(69,84) # right leg
part[5]= np.arange(84,99) # left leg
n_part = 6
for i, p in enumerate(part):
n_seg = len(p)
for j, m in enumerate(p):
COLOR_MAP[m, :] = np.array([k * 255 for k in list(cc_map(i/n_part + (1/n_part) * (j/n_seg))[:3])])
self.COLOR_MAPS["default"] = copy.deepcopy(COLOR_MAP)
## legacy
COLOR_MAP = []
palette = [[0,0,1], [0,1,0], [1,0,0], [0,1,1], [1,0,1], [1,1,0],\
[1,0,0.5], [0,1,0.5], [1,0.5,0], [0.5, 0,1],[0,0.5,1], [0.5,1,0], [0.7,0.7,0.7], \
[0.7,0.3,0], [0.3, 0,0.7], [0,0.3,0.7], [0.3,0.7,0], [0.7,0,0.3], [0,0.7,0.3]]
scale = 5
decimator = 3
for l in range(110):
s = l % 2 + 1
r = l % len(palette)
COLOR_MAP.append((100 + s * 50) * np.array(palette[r]))
COLOR_MAP.append([0, 0, 0])
COLOR_MAP = np.array(COLOR_MAP)
self.COLOR_MAPS["legacy"] = copy.deepcopy(COLOR_MAP)