Update ⓘ_Introduction.py
Browse files- ⓘ_Introduction.py +10 -890
ⓘ_Introduction.py
CHANGED
@@ -1,24 +1,5 @@
|
|
1 |
-
from streamlit import session_state as session
|
2 |
-
import shutil
|
3 |
-
|
4 |
-
import os
|
5 |
-
import numpy as np
|
6 |
-
from sklearn import neighbors
|
7 |
-
from scipy.spatial import distance_matrix
|
8 |
-
from pygco import cut_from_graph
|
9 |
-
import streamlit_ext as ste
|
10 |
-
import open3d as o3d
|
11 |
-
import matplotlib.pyplot as plt
|
12 |
-
import matplotlib.colors as mcolors
|
13 |
-
from stqdm import stqdm
|
14 |
-
import json
|
15 |
-
from stpyvista import stpyvista
|
16 |
-
import torch
|
17 |
-
import torch.nn as nn
|
18 |
-
from torch.autograd import Variable
|
19 |
-
import torch.nn.functional as F
|
20 |
import streamlit as st
|
21 |
-
import
|
22 |
|
23 |
from PIL import Image
|
24 |
|
@@ -27,7 +8,7 @@ class TeethApp:
|
|
27 |
# Font
|
28 |
with open("utils/style.css") as css:
|
29 |
st.markdown(f"<style>{css.read()}</style>", unsafe_allow_html=True)
|
30 |
-
|
31 |
# Logo
|
32 |
self.image_path = "utils/teeth-295404_1280.png"
|
33 |
self.image = Image.open(self.image_path)
|
@@ -49,881 +30,20 @@ class TeethApp:
|
|
49 |
unsafe_allow_html=True,
|
50 |
)
|
51 |
|
52 |
-
|
53 |
-
class STN3d(nn.Module):
|
54 |
-
def __init__(self, channel):
|
55 |
-
super(STN3d, self).__init__()
|
56 |
-
self.conv1 = torch.nn.Conv1d(channel, 64, 1)
|
57 |
-
self.conv2 = torch.nn.Conv1d(64, 128, 1)
|
58 |
-
self.conv3 = torch.nn.Conv1d(128, 1024, 1)
|
59 |
-
self.fc1 = nn.Linear(1024, 512)
|
60 |
-
self.fc2 = nn.Linear(512, 256)
|
61 |
-
self.fc3 = nn.Linear(256, 9)
|
62 |
-
self.relu = nn.ReLU()
|
63 |
-
|
64 |
-
self.bn1 = nn.BatchNorm1d(64)
|
65 |
-
self.bn2 = nn.BatchNorm1d(128)
|
66 |
-
self.bn3 = nn.BatchNorm1d(1024)
|
67 |
-
self.bn4 = nn.BatchNorm1d(512)
|
68 |
-
self.bn5 = nn.BatchNorm1d(256)
|
69 |
-
|
70 |
-
def forward(self, x):
|
71 |
-
batchsize = x.size()[0]
|
72 |
-
x = F.relu(self.bn1(self.conv1(x)))
|
73 |
-
x = F.relu(self.bn2(self.conv2(x)))
|
74 |
-
x = F.relu(self.bn3(self.conv3(x)))
|
75 |
-
x = torch.max(x, 2, keepdim=True)[0]
|
76 |
-
x = x.view(-1, 1024)
|
77 |
-
|
78 |
-
x = F.relu(self.bn4(self.fc1(x)))
|
79 |
-
x = F.relu(self.bn5(self.fc2(x)))
|
80 |
-
x = self.fc3(x)
|
81 |
-
|
82 |
-
iden = Variable(torch.from_numpy(np.array([1, 0, 0, 0, 1, 0, 0, 0, 1]).astype(np.float32))).view(1, 9).repeat(
|
83 |
-
batchsize, 1)
|
84 |
-
if x.is_cuda:
|
85 |
-
iden = iden.to(x.get_device())
|
86 |
-
x = x + iden
|
87 |
-
x = x.view(-1, 3, 3)
|
88 |
-
return x
|
89 |
-
|
90 |
-
class STNkd(nn.Module):
|
91 |
-
def __init__(self, k=64):
|
92 |
-
super(STNkd, self).__init__()
|
93 |
-
self.conv1 = torch.nn.Conv1d(k, 64, 1)
|
94 |
-
self.conv2 = torch.nn.Conv1d(64, 128, 1)
|
95 |
-
self.conv3 = torch.nn.Conv1d(128, 512, 1)
|
96 |
-
self.fc1 = nn.Linear(512, 256)
|
97 |
-
self.fc2 = nn.Linear(256, 128)
|
98 |
-
self.fc3 = nn.Linear(128, k * k)
|
99 |
-
self.relu = nn.ReLU()
|
100 |
-
|
101 |
-
self.bn1 = nn.BatchNorm1d(64)
|
102 |
-
self.bn2 = nn.BatchNorm1d(128)
|
103 |
-
self.bn3 = nn.BatchNorm1d(512)
|
104 |
-
self.bn4 = nn.BatchNorm1d(256)
|
105 |
-
self.bn5 = nn.BatchNorm1d(128)
|
106 |
-
|
107 |
-
self.k = k
|
108 |
-
|
109 |
-
def forward(self, x):
|
110 |
-
batchsize = x.size()[0]
|
111 |
-
x = F.relu(self.bn1(self.conv1(x)))
|
112 |
-
x = F.relu(self.bn2(self.conv2(x)))
|
113 |
-
x = F.relu(self.bn3(self.conv3(x)))
|
114 |
-
x = torch.max(x, 2, keepdim=True)[0]
|
115 |
-
x = x.view(-1, 512)
|
116 |
-
|
117 |
-
x = F.relu(self.bn4(self.fc1(x)))
|
118 |
-
x = F.relu(self.bn5(self.fc2(x)))
|
119 |
-
x = self.fc3(x)
|
120 |
-
|
121 |
-
iden = Variable(torch.from_numpy(np.eye(self.k).flatten().astype(np.float32))).view(1, self.k * self.k).repeat(
|
122 |
-
batchsize, 1)
|
123 |
-
if x.is_cuda:
|
124 |
-
iden = iden.to(x.get_device())
|
125 |
-
x = x + iden
|
126 |
-
x = x.view(-1, self.k, self.k)
|
127 |
-
return x
|
128 |
-
|
129 |
-
class MeshSegNet(nn.Module):
|
130 |
-
def __init__(self, num_classes=17, num_channels=15, with_dropout=True, dropout_p=0.5):
|
131 |
-
super(MeshSegNet, self).__init__()
|
132 |
-
self.num_classes = num_classes
|
133 |
-
self.num_channels = num_channels
|
134 |
-
self.with_dropout = with_dropout
|
135 |
-
self.dropout_p = dropout_p
|
136 |
-
|
137 |
-
# MLP-1 [64, 64]
|
138 |
-
self.mlp1_conv1 = torch.nn.Conv1d(self.num_channels, 64, 1)
|
139 |
-
self.mlp1_conv2 = torch.nn.Conv1d(64, 64, 1)
|
140 |
-
self.mlp1_bn1 = nn.BatchNorm1d(64)
|
141 |
-
self.mlp1_bn2 = nn.BatchNorm1d(64)
|
142 |
-
# FTM (feature-transformer module)
|
143 |
-
self.fstn = STNkd(k=64)
|
144 |
-
# GLM-1 (graph-contrained learning modulus)
|
145 |
-
self.glm1_conv1_1 = torch.nn.Conv1d(64, 32, 1)
|
146 |
-
self.glm1_conv1_2 = torch.nn.Conv1d(64, 32, 1)
|
147 |
-
self.glm1_bn1_1 = nn.BatchNorm1d(32)
|
148 |
-
self.glm1_bn1_2 = nn.BatchNorm1d(32)
|
149 |
-
self.glm1_conv2 = torch.nn.Conv1d(32+32, 64, 1)
|
150 |
-
self.glm1_bn2 = nn.BatchNorm1d(64)
|
151 |
-
# MLP-2
|
152 |
-
self.mlp2_conv1 = torch.nn.Conv1d(64, 64, 1)
|
153 |
-
self.mlp2_bn1 = nn.BatchNorm1d(64)
|
154 |
-
self.mlp2_conv2 = torch.nn.Conv1d(64, 128, 1)
|
155 |
-
self.mlp2_bn2 = nn.BatchNorm1d(128)
|
156 |
-
self.mlp2_conv3 = torch.nn.Conv1d(128, 512, 1)
|
157 |
-
self.mlp2_bn3 = nn.BatchNorm1d(512)
|
158 |
-
# GLM-2 (graph-contrained learning modulus)
|
159 |
-
self.glm2_conv1_1 = torch.nn.Conv1d(512, 128, 1)
|
160 |
-
self.glm2_conv1_2 = torch.nn.Conv1d(512, 128, 1)
|
161 |
-
self.glm2_conv1_3 = torch.nn.Conv1d(512, 128, 1)
|
162 |
-
self.glm2_bn1_1 = nn.BatchNorm1d(128)
|
163 |
-
self.glm2_bn1_2 = nn.BatchNorm1d(128)
|
164 |
-
self.glm2_bn1_3 = nn.BatchNorm1d(128)
|
165 |
-
self.glm2_conv2 = torch.nn.Conv1d(128*3, 512, 1)
|
166 |
-
self.glm2_bn2 = nn.BatchNorm1d(512)
|
167 |
-
# MLP-3
|
168 |
-
self.mlp3_conv1 = torch.nn.Conv1d(64+512+512+512, 256, 1)
|
169 |
-
self.mlp3_conv2 = torch.nn.Conv1d(256, 256, 1)
|
170 |
-
self.mlp3_bn1_1 = nn.BatchNorm1d(256)
|
171 |
-
self.mlp3_bn1_2 = nn.BatchNorm1d(256)
|
172 |
-
self.mlp3_conv3 = torch.nn.Conv1d(256, 128, 1)
|
173 |
-
self.mlp3_conv4 = torch.nn.Conv1d(128, 128, 1)
|
174 |
-
self.mlp3_bn2_1 = nn.BatchNorm1d(128)
|
175 |
-
self.mlp3_bn2_2 = nn.BatchNorm1d(128)
|
176 |
-
# output
|
177 |
-
self.output_conv = torch.nn.Conv1d(128, self.num_classes, 1)
|
178 |
-
if self.with_dropout:
|
179 |
-
self.dropout = nn.Dropout(p=self.dropout_p)
|
180 |
-
|
181 |
-
def forward(self, x, a_s, a_l):
|
182 |
-
batchsize = x.size()[0]
|
183 |
-
n_pts = x.size()[2]
|
184 |
-
# MLP-1
|
185 |
-
x = F.relu(self.mlp1_bn1(self.mlp1_conv1(x)))
|
186 |
-
x = F.relu(self.mlp1_bn2(self.mlp1_conv2(x)))
|
187 |
-
# FTM
|
188 |
-
trans_feat = self.fstn(x)
|
189 |
-
x = x.transpose(2, 1)
|
190 |
-
x_ftm = torch.bmm(x, trans_feat)
|
191 |
-
# GLM-1
|
192 |
-
sap = torch.bmm(a_s, x_ftm)
|
193 |
-
sap = sap.transpose(2, 1)
|
194 |
-
x_ftm = x_ftm.transpose(2, 1)
|
195 |
-
x = F.relu(self.glm1_bn1_1(self.glm1_conv1_1(x_ftm)))
|
196 |
-
glm_1_sap = F.relu(self.glm1_bn1_2(self.glm1_conv1_2(sap)))
|
197 |
-
x = torch.cat([x, glm_1_sap], dim=1)
|
198 |
-
x = F.relu(self.glm1_bn2(self.glm1_conv2(x)))
|
199 |
-
# MLP-2
|
200 |
-
x = F.relu(self.mlp2_bn1(self.mlp2_conv1(x)))
|
201 |
-
x = F.relu(self.mlp2_bn2(self.mlp2_conv2(x)))
|
202 |
-
x_mlp2 = F.relu(self.mlp2_bn3(self.mlp2_conv3(x)))
|
203 |
-
if self.with_dropout:
|
204 |
-
x_mlp2 = self.dropout(x_mlp2)
|
205 |
-
# GLM-2
|
206 |
-
x_mlp2 = x_mlp2.transpose(2, 1)
|
207 |
-
sap_1 = torch.bmm(a_s, x_mlp2)
|
208 |
-
sap_2 = torch.bmm(a_l, x_mlp2)
|
209 |
-
x_mlp2 = x_mlp2.transpose(2, 1)
|
210 |
-
sap_1 = sap_1.transpose(2, 1)
|
211 |
-
sap_2 = sap_2.transpose(2, 1)
|
212 |
-
x = F.relu(self.glm2_bn1_1(self.glm2_conv1_1(x_mlp2)))
|
213 |
-
glm_2_sap_1 = F.relu(self.glm2_bn1_2(self.glm2_conv1_2(sap_1)))
|
214 |
-
glm_2_sap_2 = F.relu(self.glm2_bn1_3(self.glm2_conv1_3(sap_2)))
|
215 |
-
x = torch.cat([x, glm_2_sap_1, glm_2_sap_2], dim=1)
|
216 |
-
x_glm2 = F.relu(self.glm2_bn2(self.glm2_conv2(x)))
|
217 |
-
# GMP
|
218 |
-
x = torch.max(x_glm2, 2, keepdim=True)[0]
|
219 |
-
# Upsample
|
220 |
-
x = torch.nn.Upsample(n_pts)(x)
|
221 |
-
# Dense fusion
|
222 |
-
x = torch.cat([x, x_ftm, x_mlp2, x_glm2], dim=1)
|
223 |
-
# MLP-3
|
224 |
-
x = F.relu(self.mlp3_bn1_1(self.mlp3_conv1(x)))
|
225 |
-
x = F.relu(self.mlp3_bn1_2(self.mlp3_conv2(x)))
|
226 |
-
x = F.relu(self.mlp3_bn2_1(self.mlp3_conv3(x)))
|
227 |
-
if self.with_dropout:
|
228 |
-
x = self.dropout(x)
|
229 |
-
x = F.relu(self.mlp3_bn2_2(self.mlp3_conv4(x)))
|
230 |
-
# output
|
231 |
-
x = self.output_conv(x)
|
232 |
-
x = x.transpose(2,1).contiguous()
|
233 |
-
x = torch.nn.Softmax(dim=-1)(x.view(-1, self.num_classes))
|
234 |
-
x = x.view(batchsize, n_pts, self.num_classes)
|
235 |
-
|
236 |
-
return x
|
237 |
-
|
238 |
-
def clone_runoob(li1):
|
239 |
-
li_copy = li1[:]
|
240 |
-
return li_copy
|
241 |
-
|
242 |
-
# 对离群点重新进行分类
|
243 |
-
def class_inlier_outlier(label_list, mean_points,cloud, ind, label_index, points, labels):
|
244 |
-
label_change = clone_runoob(labels)
|
245 |
-
outlier_index = clone_runoob(label_index)
|
246 |
-
ind_reverse = clone_runoob(ind)
|
247 |
-
# 得到离群点的label下标
|
248 |
-
ind_reverse.reverse()
|
249 |
-
for i in ind_reverse:
|
250 |
-
outlier_index.pop(i)
|
251 |
-
|
252 |
-
# 获取离群点
|
253 |
-
inlier_cloud = cloud.select_by_index(ind)
|
254 |
-
outlier_cloud = cloud.select_by_index(ind, invert=True)
|
255 |
-
outlier_points = np.array(outlier_cloud.points)
|
256 |
-
|
257 |
-
for i in range(len(outlier_points)):
|
258 |
-
distance = []
|
259 |
-
for j in range(len(mean_points)):
|
260 |
-
dis = np.linalg.norm(outlier_points[i] - mean_points[j], ord=2) # 计算tooth和GT质心之间的距离
|
261 |
-
distance.append(dis)
|
262 |
-
min_index = distance.index(min(distance)) # 获取和离群点质心最近label的index
|
263 |
-
outlier_label = label_list[min_index] # 获取离群点应该的label
|
264 |
-
index = outlier_index[i]
|
265 |
-
label_change[index] = outlier_label
|
266 |
-
|
267 |
-
return label_change
|
268 |
-
|
269 |
-
# 利用knn算法消除离群点
|
270 |
-
def remove_outlier(points, labels):
|
271 |
-
# points = np.array(point_cloud_o3d_orign.points)
|
272 |
-
# global label_list
|
273 |
-
same_label_points = {}
|
274 |
-
|
275 |
-
same_label_index = {}
|
276 |
-
|
277 |
-
mean_points = [] # 所有label种类对应点云的质心坐标
|
278 |
-
|
279 |
-
label_list = []
|
280 |
-
for i in range(len(labels)):
|
281 |
-
label_list.append(labels[i])
|
282 |
-
label_list = list(set(label_list)) # 去重获从小到大排序取GT_label=[0, 11, 12, 13, 14, 15, 16, 17, 21, 22, 23, 24, 25, 26, 27]
|
283 |
-
label_list.sort()
|
284 |
-
label_list = label_list[1:]
|
285 |
-
|
286 |
-
for i in label_list:
|
287 |
-
key = i
|
288 |
-
points_list = []
|
289 |
-
all_label_index = []
|
290 |
-
for j in range(len(labels)):
|
291 |
-
if labels[j] == i:
|
292 |
-
points_list.append(points[j].tolist())
|
293 |
-
all_label_index.append(j) # 得到label为 i 的点对应的label的下标
|
294 |
-
same_label_points[key] = points_list
|
295 |
-
same_label_index[key] = all_label_index
|
296 |
-
|
297 |
-
tooth_mean = np.mean(points_list, axis=0)
|
298 |
-
mean_points.append(tooth_mean)
|
299 |
-
# print(mean_points)
|
300 |
-
|
301 |
-
for i in label_list:
|
302 |
-
points_array = same_label_points[i]
|
303 |
-
# 建立一个o3d的点云对象
|
304 |
-
pcd = o3d.geometry.PointCloud()
|
305 |
-
# 使用Vector3dVector方法转换
|
306 |
-
pcd.points = o3d.utility.Vector3dVector(points_array)
|
307 |
-
|
308 |
-
# 对label i 对应的点云进行统计离群值去除,找出离群点并显示
|
309 |
-
# 统计式离群点移除
|
310 |
-
cl, ind = pcd.remove_statistical_outlier(nb_neighbors=200, std_ratio=2.0) # cl是选中的点,ind是选中点index
|
311 |
-
# 可视化
|
312 |
-
# display_inlier_outlier(pcd, ind)
|
313 |
-
|
314 |
-
# 对分出来的离群点重新分类
|
315 |
-
label_index = same_label_index[i]
|
316 |
-
labels = class_inlier_outlier(label_list, mean_points, pcd, ind, label_index, points, labels)
|
317 |
-
# print(f"label_change{labels[4400]}")
|
318 |
-
|
319 |
-
return labels
|
320 |
-
|
321 |
-
|
322 |
-
# 消除离群点,保存最后的输出
|
323 |
-
def remove_outlier_main(jaw, pcd_points, labels, instances_labels):
|
324 |
-
# point_cloud_o3d_orign = o3d.io.read_point_cloud('E:/tooth/data/MeshSegNet-master/test_upsample_15/upsample_01K17AN8_upper_refined.pcd')
|
325 |
-
# 原始点
|
326 |
-
points = pcd_points.copy()
|
327 |
-
label = remove_outlier(points, labels)
|
328 |
-
|
329 |
-
# 保存json文件
|
330 |
-
label_dict = {}
|
331 |
-
label_dict["id_patient"] = ""
|
332 |
-
label_dict["jaw"] = jaw
|
333 |
-
label_dict["labels"] = label.tolist()
|
334 |
-
label_dict["instances"] = instances_labels.tolist()
|
335 |
-
b = json.dumps(label_dict)
|
336 |
-
with open('dental-labels4' + '.json', 'w') as f_obj:
|
337 |
-
f_obj.write(b)
|
338 |
-
f_obj.close()
|
339 |
-
|
340 |
-
|
341 |
-
same_points_list = {}
|
342 |
-
|
343 |
-
|
344 |
-
# 体素下采样
|
345 |
-
def voxel_filter(point_cloud, leaf_size):
|
346 |
-
same_points_list = {}
|
347 |
-
filtered_points = []
|
348 |
-
# step1 计算边界点
|
349 |
-
x_max, y_max, z_max = np.amax(point_cloud, axis=0) # 计算 x,y,z三个维度的最值
|
350 |
-
x_min, y_min, z_min = np.amin(point_cloud, axis=0)
|
351 |
-
|
352 |
-
# step2 确定体素的尺寸
|
353 |
-
size_r = leaf_size
|
354 |
-
|
355 |
-
# step3 计算每个 volex的维度 voxel grid
|
356 |
-
Dx = (x_max - x_min) // size_r + 1
|
357 |
-
Dy = (y_max - y_min) // size_r + 1
|
358 |
-
Dz = (z_max - z_min) // size_r + 1
|
359 |
-
|
360 |
-
# print("Dx x Dy x Dz is {} x {} x {}".format(Dx, Dy, Dz))
|
361 |
-
|
362 |
-
# step4 计算每个点在volex grid内每一个维度的值
|
363 |
-
h = list() # h 为保存索引的列表
|
364 |
-
for i in range(len(point_cloud)):
|
365 |
-
hx = np.floor((point_cloud[i][0] - x_min) // size_r)
|
366 |
-
hy = np.floor((point_cloud[i][1] - y_min) // size_r)
|
367 |
-
hz = np.floor((point_cloud[i][2] - z_min) // size_r)
|
368 |
-
h.append(hx + hy * Dx + hz * Dx * Dy)
|
369 |
-
# print(h[60581])
|
370 |
-
|
371 |
-
# step5 对h值进行排序
|
372 |
-
h = np.array(h)
|
373 |
-
h_indice = np.argsort(h) # 提取索引,返回h里面的元素按从小到大排序的 索引
|
374 |
-
h_sorted = h[h_indice] # 升序
|
375 |
-
count = 0 # 用于维度的累计
|
376 |
-
step = 20
|
377 |
-
# 将h值相同的点放入到同一个grid中,并进行筛选
|
378 |
-
for i in range(1, len(h_sorted)): # 0-19999个数据点
|
379 |
-
# if i == len(h_sorted)-1:
|
380 |
-
# print("aaa")
|
381 |
-
if h_sorted[i] == h_sorted[i - 1] and (i != len(h_sorted) - 1):
|
382 |
-
continue
|
383 |
-
elif h_sorted[i] == h_sorted[i - 1] and (i == len(h_sorted) - 1):
|
384 |
-
point_idx = h_indice[count:]
|
385 |
-
key = h_sorted[i - 1]
|
386 |
-
same_points_list[key] = point_idx
|
387 |
-
_G = np.mean(point_cloud[point_idx], axis=0) # 所有点的重心
|
388 |
-
_d = np.linalg.norm(point_cloud[point_idx] - _G, axis=1, ord=2) # 计算到重心的距离
|
389 |
-
_d.sort()
|
390 |
-
inx = [j for j in range(0, len(_d), step)] # 获取指定间隔元素下标
|
391 |
-
for j in inx:
|
392 |
-
index = point_idx[j]
|
393 |
-
filtered_points.append(point_cloud[index])
|
394 |
-
count = i
|
395 |
-
elif h_sorted[i] != h_sorted[i - 1] and (i == len(h_sorted) - 1):
|
396 |
-
point_idx1 = h_indice[count:i]
|
397 |
-
key1 = h_sorted[i - 1]
|
398 |
-
same_points_list[key1] = point_idx1
|
399 |
-
_G = np.mean(point_cloud[point_idx1], axis=0) # 所有点的重心
|
400 |
-
_d = np.linalg.norm(point_cloud[point_idx1] - _G, axis=1, ord=2) # 计算到重心的距离
|
401 |
-
_d.sort()
|
402 |
-
inx = [j for j in range(0, len(_d), step)] # 获取指定间隔元素下标
|
403 |
-
for j in inx:
|
404 |
-
index = point_idx1[j]
|
405 |
-
filtered_points.append(point_cloud[index])
|
406 |
-
|
407 |
-
point_idx2 = h_indice[i:]
|
408 |
-
key2 = h_sorted[i]
|
409 |
-
same_points_list[key2] = point_idx2
|
410 |
-
_G = np.mean(point_cloud[point_idx2], axis=0) # 所有点的重心
|
411 |
-
_d = np.linalg.norm(point_cloud[point_idx2] - _G, axis=1, ord=2) # 计算到重心的距离
|
412 |
-
_d.sort()
|
413 |
-
inx = [j for j in range(0, len(_d), step)] # 获取指定间隔元素下标
|
414 |
-
for j in inx:
|
415 |
-
index = point_idx2[j]
|
416 |
-
filtered_points.append(point_cloud[index])
|
417 |
-
count = i
|
418 |
-
|
419 |
-
else:
|
420 |
-
point_idx = h_indice[count: i]
|
421 |
-
key = h_sorted[i - 1]
|
422 |
-
same_points_list[key] = point_idx
|
423 |
-
_G = np.mean(point_cloud[point_idx], axis=0) # 所有点的重心
|
424 |
-
_d = np.linalg.norm(point_cloud[point_idx] - _G, axis=1, ord=2) # 计算到重心的距离
|
425 |
-
_d.sort()
|
426 |
-
inx = [j for j in range(0, len(_d), step)] # 获取指定间隔元素下标
|
427 |
-
for j in inx:
|
428 |
-
index = point_idx[j]
|
429 |
-
filtered_points.append(point_cloud[index])
|
430 |
-
count = i
|
431 |
-
|
432 |
-
# 把点云格式改成array,并对外返回
|
433 |
-
# print(f'filtered_points[0]为{filtered_points[0]}')
|
434 |
-
filtered_points = np.array(filtered_points, dtype=np.float64)
|
435 |
-
return filtered_points,same_points_list
|
436 |
-
|
437 |
-
|
438 |
-
# 体素上采样
|
439 |
-
def voxel_upsample(same_points_list, point_cloud, filtered_points, filter_labels, leaf_size):
|
440 |
-
upsample_label = []
|
441 |
-
upsample_point = []
|
442 |
-
upsample_index = []
|
443 |
-
# step1 计算边界点
|
444 |
-
x_max, y_max, z_max = np.amax(point_cloud, axis=0) # 计算 x,y,z三个维度的最值
|
445 |
-
x_min, y_min, z_min = np.amin(point_cloud, axis=0)
|
446 |
-
# step2 确定体素的尺寸
|
447 |
-
size_r = leaf_size
|
448 |
-
# step3 计算每个 volex的维度 voxel grid
|
449 |
-
Dx = (x_max - x_min) // size_r + 1
|
450 |
-
Dy = (y_max - y_min) // size_r + 1
|
451 |
-
Dz = (z_max - z_min) // size_r + 1
|
452 |
-
print("Dx x Dy x Dz is {} x {} x {}".format(Dx, Dy, Dz))
|
453 |
-
|
454 |
-
# step4 计算每个点(采样后的点)在volex grid内每一个维度的值
|
455 |
-
h = list()
|
456 |
-
for i in range(len(filtered_points)):
|
457 |
-
hx = np.floor((filtered_points[i][0] - x_min) // size_r)
|
458 |
-
hy = np.floor((filtered_points[i][1] - y_min) // size_r)
|
459 |
-
hz = np.floor((filtered_points[i][2] - z_min) // size_r)
|
460 |
-
h.append(hx + hy * Dx + hz * Dx * Dy)
|
461 |
-
|
462 |
-
# step5 根据h值查询字典same_points_list
|
463 |
-
h = np.array(h)
|
464 |
-
count = 0
|
465 |
-
for i in range(1, len(h)):
|
466 |
-
if h[i] == h[i - 1] and i != (len(h) - 1):
|
467 |
-
continue
|
468 |
-
elif h[i] == h[i - 1] and i == (len(h) - 1):
|
469 |
-
label = filter_labels[count:]
|
470 |
-
key = h[i - 1]
|
471 |
-
count = i
|
472 |
-
# 累计label次数,classcount:{‘A’:2,'B':1}
|
473 |
-
classcount = {}
|
474 |
-
for i in range(len(label)):
|
475 |
-
vote = label[i]
|
476 |
-
classcount[vote] = classcount.get(vote, 0) + 1
|
477 |
-
# 对map的value排序
|
478 |
-
sortedclass = sorted(classcount.items(), key=lambda x: (x[1]), reverse=True)
|
479 |
-
# key = h[i-1]
|
480 |
-
point_index = same_points_list[key] # h对应的point index列表
|
481 |
-
for j in range(len(point_index)):
|
482 |
-
upsample_label.append(sortedclass[0][0])
|
483 |
-
index = point_index[j]
|
484 |
-
upsample_point.append(point_cloud[index])
|
485 |
-
upsample_index.append(index)
|
486 |
-
elif h[i] != h[i - 1] and (i == len(h) - 1):
|
487 |
-
label1 = filter_labels[count:i]
|
488 |
-
key1 = h[i - 1]
|
489 |
-
label2 = filter_labels[i:]
|
490 |
-
key2 = h[i]
|
491 |
-
count = i
|
492 |
-
|
493 |
-
classcount = {}
|
494 |
-
for i in range(len(label1)):
|
495 |
-
vote = label1[i]
|
496 |
-
classcount[vote] = classcount.get(vote, 0) + 1
|
497 |
-
sortedclass = sorted(classcount.items(), key=lambda x: (x[1]), reverse=True)
|
498 |
-
# key1 = h[i-1]
|
499 |
-
point_index = same_points_list[key1]
|
500 |
-
for j in range(len(point_index)):
|
501 |
-
upsample_label.append(sortedclass[0][0])
|
502 |
-
index = point_index[j]
|
503 |
-
upsample_point.append(point_cloud[index])
|
504 |
-
upsample_index.append(index)
|
505 |
-
|
506 |
-
# label2 = filter_labels[i:]
|
507 |
-
classcount = {}
|
508 |
-
for i in range(len(label2)):
|
509 |
-
vote = label2[i]
|
510 |
-
classcount[vote] = classcount.get(vote, 0) + 1
|
511 |
-
sortedclass = sorted(classcount.items(), key=lambda x: (x[1]), reverse=True)
|
512 |
-
# key2 = h[i]
|
513 |
-
point_index = same_points_list[key2]
|
514 |
-
for j in range(len(point_index)):
|
515 |
-
upsample_label.append(sortedclass[0][0])
|
516 |
-
index = point_index[j]
|
517 |
-
upsample_point.append(point_cloud[index])
|
518 |
-
upsample_index.append(index)
|
519 |
-
else:
|
520 |
-
label = filter_labels[count:i]
|
521 |
-
key = h[i - 1]
|
522 |
-
count = i
|
523 |
-
classcount = {}
|
524 |
-
for i in range(len(label)):
|
525 |
-
vote = label[i]
|
526 |
-
classcount[vote] = classcount.get(vote, 0) + 1
|
527 |
-
sortedclass = sorted(classcount.items(), key=lambda x: (x[1]), reverse=True)
|
528 |
-
# key = h[i-1]
|
529 |
-
point_index = same_points_list[key] # h对应的point index列表
|
530 |
-
for j in range(len(point_index)):
|
531 |
-
upsample_label.append(sortedclass[0][0])
|
532 |
-
index = point_index[j]
|
533 |
-
upsample_point.append(point_cloud[index])
|
534 |
-
upsample_index.append(index)
|
535 |
-
# count = i
|
536 |
-
|
537 |
-
# 恢复原始顺序
|
538 |
-
# print(f'upsample_index[0]的值为{upsample_index[0]}')
|
539 |
-
# print(f'upsample_index的总长度为{len(upsample_index)}')
|
540 |
-
|
541 |
-
# 恢复index原始顺序
|
542 |
-
upsample_index = np.array(upsample_index)
|
543 |
-
upsample_index_indice = np.argsort(upsample_index) # 提取索引,返回h里面的元素按从小到大排序的 索引
|
544 |
-
upsample_index_sorted = upsample_index[upsample_index_indice]
|
545 |
-
|
546 |
-
upsample_point = np.array(upsample_point)
|
547 |
-
upsample_label = np.array(upsample_label)
|
548 |
-
# 恢复point和label的原始顺序
|
549 |
-
upsample_point_sorted = upsample_point[upsample_index_indice]
|
550 |
-
upsample_label_sorted = upsample_label[upsample_index_indice]
|
551 |
-
|
552 |
-
return upsample_point_sorted, upsample_label_sorted
|
553 |
-
|
554 |
-
|
555 |
-
# 利用knn算法上采样
|
556 |
-
def KNN_sklearn_Load_data(voxel_points, center_points, labels):
|
557 |
-
# 载入数据
|
558 |
-
# x_train, x_test, y_train, y_test = train_test_split(center_points, labels, test_size=0.1)
|
559 |
-
# 构建模型
|
560 |
-
model = neighbors.KNeighborsClassifier(n_neighbors=3)
|
561 |
-
model.fit(center_points, labels)
|
562 |
-
prediction = model.predict(voxel_points.reshape(1, -1))
|
563 |
-
# meshtopoints_labels = classification_report(voxel_points, prediction)
|
564 |
-
return prediction[0]
|
565 |
-
|
566 |
-
|
567 |
-
# 加载点进行knn上采样
|
568 |
-
def Load_data(voxel_points, center_points, labels):
|
569 |
-
meshtopoints_labels = []
|
570 |
-
# meshtopoints_labels.append(SVC_sklearn_Load_data(voxel_points[i], center_points, labels))
|
571 |
-
for i in range(0, voxel_points.shape[0]):
|
572 |
-
meshtopoints_labels.append(KNN_sklearn_Load_data(voxel_points[i], center_points, labels))
|
573 |
-
return np.array(meshtopoints_labels)
|
574 |
-
|
575 |
-
# 将三角网格数据上采样回原始点云数据
|
576 |
-
def mesh_to_points_main(jaw, pcd_points, center_points, labels):
|
577 |
-
points = pcd_points.copy()
|
578 |
-
# 下采样
|
579 |
-
voxel_points, same_points_list = voxel_filter(points, 0.6)
|
580 |
-
|
581 |
-
after_labels = Load_data(voxel_points, center_points, labels)
|
582 |
-
|
583 |
-
upsample_point, upsample_label = voxel_upsample(same_points_list, points, voxel_points, after_labels, 0.6)
|
584 |
-
|
585 |
-
new_pcd = o3d.geometry.PointCloud()
|
586 |
-
new_pcd.points = o3d.utility.Vector3dVector(upsample_point)
|
587 |
-
instances_labels = upsample_label.copy()
|
588 |
-
# '''
|
589 |
-
# o3d.io.write_point_cloud(os.path.join(save_path, 'upsample_' + name + '.pcd'), new_pcd, write_ascii=True)
|
590 |
-
for i in stqdm(range(0, upsample_label.shape[0])):
|
591 |
-
if jaw == 'upper':
|
592 |
-
if (upsample_label[i] >= 1) and (upsample_label[i] <= 8):
|
593 |
-
upsample_label[i] = upsample_label[i] + 10
|
594 |
-
elif (upsample_label[i] >= 9) and (upsample_label[i] <= 16):
|
595 |
-
upsample_label[i] = upsample_label[i] + 12
|
596 |
-
else:
|
597 |
-
if (upsample_label[i] >= 1) and (upsample_label[i] <= 8):
|
598 |
-
upsample_label[i] = upsample_label[i] + 30
|
599 |
-
elif (upsample_label[i] >= 9) and (upsample_label[i] <= 16):
|
600 |
-
upsample_label[i] = upsample_label[i] + 32
|
601 |
-
remove_outlier_main(jaw, pcd_points, upsample_label, instances_labels)
|
602 |
-
|
603 |
-
|
604 |
-
# 将原始点云数据转换为三角网格
|
605 |
-
def mesh_grid(pcd_points):
|
606 |
-
new_pcd,_ = voxel_filter(pcd_points, 0.6)
|
607 |
-
# pcd需要有法向量
|
608 |
-
|
609 |
-
# estimate radius for rolling ball
|
610 |
-
pcd_new = o3d.geometry.PointCloud()
|
611 |
-
pcd_new.points = o3d.utility.Vector3dVector(new_pcd)
|
612 |
-
pcd_new.estimate_normals()
|
613 |
-
distances = pcd_new.compute_nearest_neighbor_distance()
|
614 |
-
avg_dist = np.mean(distances)
|
615 |
-
radius = 6 * avg_dist
|
616 |
-
mesh = o3d.geometry.TriangleMesh.create_from_point_cloud_ball_pivoting(
|
617 |
-
pcd_new,
|
618 |
-
o3d.utility.DoubleVector([radius, radius * 2]))
|
619 |
-
# o3d.io.write_triangle_mesh("./tooth date/test.ply", mesh)
|
620 |
-
|
621 |
-
return mesh
|
622 |
-
|
623 |
-
|
624 |
-
# 读取obj文件内容
|
625 |
-
def read_obj(obj_path):
|
626 |
-
jaw = None
|
627 |
-
with open(obj_path) as file:
|
628 |
-
points = []
|
629 |
-
faces = []
|
630 |
-
while 1:
|
631 |
-
line = file.readline()
|
632 |
-
if not line:
|
633 |
-
break
|
634 |
-
strs = line.split(" ")
|
635 |
-
if strs[0] == "v":
|
636 |
-
points.append((float(strs[1]), float(strs[2]), float(strs[3])))
|
637 |
-
elif strs[0] == "f":
|
638 |
-
faces.append((int(strs[1]), int(strs[2]), int(strs[3])))
|
639 |
-
elif strs[1][0:5] == 'lower':
|
640 |
-
jaw = 'lower'
|
641 |
-
elif strs[1][0:5] == 'upper':
|
642 |
-
jaw = 'upper'
|
643 |
-
|
644 |
-
points = np.array(points)
|
645 |
-
faces = np.array(faces)
|
646 |
-
|
647 |
-
if jaw is None:
|
648 |
-
raise ValueError("Jaw type not found in OBJ file")
|
649 |
-
|
650 |
-
return points, faces, jaw
|
651 |
-
|
652 |
-
|
653 |
-
# obj文件转为pcd文件
|
654 |
-
def obj2pcd(obj_path):
|
655 |
-
if os.path.exists(obj_path):
|
656 |
-
print('yes')
|
657 |
-
points, _, jaw = read_obj(obj_path)
|
658 |
-
pcd_list = []
|
659 |
-
num_points = np.shape(points)[0]
|
660 |
-
for i in range(num_points):
|
661 |
-
new_line = str(points[i, 0]) + ' ' + str(points[i, 1]) + ' ' + str(points[i, 2])
|
662 |
-
pcd_list.append(new_line.split())
|
663 |
-
|
664 |
-
pcd_points = np.array(pcd_list).astype(np.float64)
|
665 |
-
return pcd_points, jaw
|
666 |
-
|
667 |
-
|
668 |
-
def segmentation_main(obj_path):
|
669 |
-
upsampling_method = 'KNN'
|
670 |
-
|
671 |
-
model_path = 'Mesh_Segementation_MeshSegNet_17_classes_60samples_best.tar'
|
672 |
-
num_classes = 17
|
673 |
-
num_channels = 15
|
674 |
-
|
675 |
-
# set model
|
676 |
-
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
|
677 |
-
model = MeshSegNet(num_classes=num_classes, num_channels=num_channels).to(device, dtype=torch.float)
|
678 |
-
|
679 |
-
# load trained model
|
680 |
-
# checkpoint = torch.load(os.path.join(model_path, model_name), map_location='cpu')
|
681 |
-
checkpoint = torch.load(model_path, map_location='cpu')
|
682 |
-
model.load_state_dict(checkpoint['model_state_dict'])
|
683 |
-
del checkpoint
|
684 |
-
model = model.to(device, dtype=torch.float)
|
685 |
-
|
686 |
-
# cudnn
|
687 |
-
torch.backends.cudnn.benchmark = True
|
688 |
-
torch.backends.cudnn.enabled = True
|
689 |
-
|
690 |
-
# Predicting
|
691 |
-
model.eval()
|
692 |
-
with torch.no_grad():
|
693 |
-
pcd_points, jaw = obj2pcd(obj_path)
|
694 |
-
mesh = mesh_grid(pcd_points)
|
695 |
-
|
696 |
-
# move mesh to origin
|
697 |
-
with st.spinner("Patience please, AI at work. Grab a coffee while you wait ☕."):
|
698 |
-
vertices_points = np.asarray(mesh.vertices)
|
699 |
-
triangles_points = np.asarray(mesh.triangles)
|
700 |
-
N = triangles_points.shape[0]
|
701 |
-
cells = np.zeros((triangles_points.shape[0], 9))
|
702 |
-
cells = vertices_points[triangles_points].reshape(triangles_points.shape[0], 9)
|
703 |
-
|
704 |
-
mean_cell_centers = mesh.get_center()
|
705 |
-
cells[:, 0:3] -= mean_cell_centers[0:3]
|
706 |
-
cells[:, 3:6] -= mean_cell_centers[0:3]
|
707 |
-
cells[:, 6:9] -= mean_cell_centers[0:3]
|
708 |
-
|
709 |
-
v1 = np.zeros([triangles_points.shape[0], 3], dtype='float32')
|
710 |
-
v2 = np.zeros([triangles_points.shape[0], 3], dtype='float32')
|
711 |
-
v1[:, 0] = cells[:, 0] - cells[:, 3]
|
712 |
-
v1[:, 1] = cells[:, 1] - cells[:, 4]
|
713 |
-
v1[:, 2] = cells[:, 2] - cells[:, 5]
|
714 |
-
v2[:, 0] = cells[:, 3] - cells[:, 6]
|
715 |
-
v2[:, 1] = cells[:, 4] - cells[:, 7]
|
716 |
-
v2[:, 2] = cells[:, 5] - cells[:, 8]
|
717 |
-
mesh_normals = np.cross(v1, v2)
|
718 |
-
mesh_normal_length = np.linalg.norm(mesh_normals, axis=1)
|
719 |
-
mesh_normals[:, 0] /= mesh_normal_length[:]
|
720 |
-
mesh_normals[:, 1] /= mesh_normal_length[:]
|
721 |
-
mesh_normals[:, 2] /= mesh_normal_length[:]
|
722 |
-
|
723 |
-
# prepare input
|
724 |
-
points = vertices_points.copy()
|
725 |
-
points[:, 0:3] -= mean_cell_centers[0:3]
|
726 |
-
normals = np.nan_to_num(mesh_normals).copy()
|
727 |
-
barycenters = np.zeros((triangles_points.shape[0], 3))
|
728 |
-
s = np.sum(vertices_points[triangles_points], 1)
|
729 |
-
barycenters = 1 / 3 * s
|
730 |
-
center_points = barycenters.copy()
|
731 |
-
barycenters -= mean_cell_centers[0:3]
|
732 |
-
|
733 |
-
# normalized data
|
734 |
-
maxs = points.max(axis=0)
|
735 |
-
mins = points.min(axis=0)
|
736 |
-
means = points.mean(axis=0)
|
737 |
-
stds = points.std(axis=0)
|
738 |
-
nmeans = normals.mean(axis=0)
|
739 |
-
nstds = normals.std(axis=0)
|
740 |
-
|
741 |
-
for i in range(3):
|
742 |
-
cells[:, i] = (cells[:, i] - means[i]) / stds[i] # point 1
|
743 |
-
cells[:, i + 3] = (cells[:, i + 3] - means[i]) / stds[i] # point 2
|
744 |
-
cells[:, i + 6] = (cells[:, i + 6] - means[i]) / stds[i] # point 3
|
745 |
-
barycenters[:, i] = (barycenters[:, i] - mins[i]) / (maxs[i] - mins[i])
|
746 |
-
normals[:, i] = (normals[:, i] - nmeans[i]) / nstds[i]
|
747 |
-
|
748 |
-
X = np.column_stack((cells, barycenters, normals))
|
749 |
-
|
750 |
-
# computing A_S and A_L
|
751 |
-
A_S = np.zeros([X.shape[0], X.shape[0]], dtype='float32')
|
752 |
-
A_L = np.zeros([X.shape[0], X.shape[0]], dtype='float32')
|
753 |
-
D = distance_matrix(X[:, 9:12], X[:, 9:12])
|
754 |
-
A_S[D < 0.1] = 1.0
|
755 |
-
A_S = A_S / np.dot(np.sum(A_S, axis=1, keepdims=True), np.ones((1, X.shape[0])))
|
756 |
-
|
757 |
-
A_L[D < 0.2] = 1.0
|
758 |
-
A_L = A_L / np.dot(np.sum(A_L, axis=1, keepdims=True), np.ones((1, X.shape[0])))
|
759 |
-
|
760 |
-
# numpy -> torch.tensor
|
761 |
-
X = X.transpose(1, 0)
|
762 |
-
X = X.reshape([1, X.shape[0], X.shape[1]])
|
763 |
-
X = torch.from_numpy(X).to(device, dtype=torch.float)
|
764 |
-
A_S = A_S.reshape([1, A_S.shape[0], A_S.shape[1]])
|
765 |
-
A_L = A_L.reshape([1, A_L.shape[0], A_L.shape[1]])
|
766 |
-
A_S = torch.from_numpy(A_S).to(device, dtype=torch.float)
|
767 |
-
A_L = torch.from_numpy(A_L).to(device, dtype=torch.float)
|
768 |
-
|
769 |
-
tensor_prob_output = model(X, A_S, A_L).to(device, dtype=torch.float)
|
770 |
-
patch_prob_output = tensor_prob_output.cpu().numpy()
|
771 |
-
|
772 |
-
# refinement
|
773 |
-
with st.spinner("Refining..."):
|
774 |
-
round_factor = 100
|
775 |
-
patch_prob_output[patch_prob_output < 1.0e-6] = 1.0e-6
|
776 |
-
|
777 |
-
# unaries
|
778 |
-
unaries = -round_factor * np.log10(patch_prob_output)
|
779 |
-
unaries = unaries.astype(np.int32)
|
780 |
-
unaries = unaries.reshape(-1, num_classes)
|
781 |
-
|
782 |
-
# parawisex
|
783 |
-
pairwise = (1 - np.eye(num_classes, dtype=np.int32))
|
784 |
-
|
785 |
-
cells = cells.copy()
|
786 |
-
|
787 |
-
cell_ids = np.asarray(triangles_points)
|
788 |
-
lambda_c = 20
|
789 |
-
edges = np.empty([1, 3], order='C')
|
790 |
-
for i_node in stqdm(range(cells.shape[0])):
|
791 |
-
# Find neighbors
|
792 |
-
nei = np.sum(np.isin(cell_ids, cell_ids[i_node, :]), axis=1)
|
793 |
-
nei_id = np.where(nei == 2)
|
794 |
-
for i_nei in nei_id[0][:]:
|
795 |
-
if i_node < i_nei:
|
796 |
-
cos_theta = np.dot(normals[i_node, 0:3], normals[i_nei, 0:3]) / np.linalg.norm(
|
797 |
-
normals[i_node, 0:3]) / np.linalg.norm(normals[i_nei, 0:3])
|
798 |
-
if cos_theta >= 1.0:
|
799 |
-
cos_theta = 0.9999
|
800 |
-
theta = np.arccos(cos_theta)
|
801 |
-
phi = np.linalg.norm(barycenters[i_node, :] - barycenters[i_nei, :])
|
802 |
-
if theta > np.pi / 2.0:
|
803 |
-
edges = np.concatenate(
|
804 |
-
(edges, np.array([i_node, i_nei, -np.log10(theta / np.pi) * phi]).reshape(1, 3)), axis=0)
|
805 |
-
else:
|
806 |
-
beta = 1 + np.linalg.norm(np.dot(normals[i_node, 0:3], normals[i_nei, 0:3]))
|
807 |
-
edges = np.concatenate(
|
808 |
-
(edges, np.array([i_node, i_nei, -beta * np.log10(theta / np.pi) * phi]).reshape(1, 3)),
|
809 |
-
axis=0)
|
810 |
-
edges = np.delete(edges, 0, 0)
|
811 |
-
edges[:, 2] *= lambda_c * round_factor
|
812 |
-
edges = edges.astype(np.int32)
|
813 |
-
|
814 |
-
refine_labels = cut_from_graph(edges, unaries, pairwise)
|
815 |
-
refine_labels = refine_labels.reshape([-1, 1])
|
816 |
-
|
817 |
-
predicted_labels_3 = refine_labels.reshape(refine_labels.shape[0])
|
818 |
-
mesh_to_points_main(jaw, pcd_points, center_points, predicted_labels_3)
|
819 |
-
|
820 |
-
import pyvista as pv
|
821 |
-
with st.spinner("Rendering..."):
|
822 |
-
# Load the .obj file
|
823 |
-
mesh = pv.read('file.obj')
|
824 |
-
|
825 |
-
# Load the JSON file
|
826 |
-
with open('dental-labels4.json', 'r') as file:
|
827 |
-
labels_data = json.load(file)
|
828 |
-
|
829 |
-
# Assuming labels_data['labels'] is a list of labels
|
830 |
-
labels = labels_data['labels']
|
831 |
-
|
832 |
-
# Make sure the number of labels matches the number of vertices or faces
|
833 |
-
assert len(labels) == mesh.n_points or len(labels) == mesh.n_cells
|
834 |
-
|
835 |
-
# If labels correspond to vertices
|
836 |
-
if len(labels) == mesh.n_points:
|
837 |
-
mesh.point_data['Labels'] = labels
|
838 |
-
# If labels correspond to faces
|
839 |
-
elif len(labels) == mesh.n_cells:
|
840 |
-
mesh.cell_data['Labels'] = labels
|
841 |
-
|
842 |
-
# Create a pyvista plotter
|
843 |
-
plotter = pv.Plotter()
|
844 |
-
|
845 |
-
cmap = plt.cm.get_cmap('jet', 27) # Using a colormap with sufficient distinct colors
|
846 |
-
|
847 |
-
colors = cmap(np.linspace(0, 1, 27)) # Generate colors
|
848 |
-
|
849 |
-
# Convert colors to a format acceptable by PyVista
|
850 |
-
colormap = mcolors.ListedColormap(colors)
|
851 |
-
|
852 |
-
# Add the mesh to the plotter with labels as a scalar field
|
853 |
-
#plotter.add_mesh(mesh, scalars='Labels', show_scalar_bar=True, cmap='jet')
|
854 |
-
plotter.add_mesh(mesh, scalars='Labels', show_scalar_bar=True, cmap=colormap, clim=[0, 27])
|
855 |
-
|
856 |
-
# Show the plot
|
857 |
-
#plotter.show()
|
858 |
-
## Send to streamlit
|
859 |
-
with st.expander("**View Segmentation Result** - ", expanded=False):
|
860 |
-
stpyvista(plotter)
|
861 |
-
|
862 |
# Configure Streamlit page
|
863 |
-
st.set_page_config(page_title="Teeth Segmentation", page_icon="
|
864 |
|
865 |
-
class
|
866 |
def __init__(self):
|
867 |
TeethApp.__init__(self)
|
868 |
self.build_app()
|
869 |
|
870 |
def build_app(self):
|
871 |
-
|
872 |
-
st.
|
873 |
-
st.markdown("
|
874 |
-
|
875 |
-
|
876 |
-
"Select scan for segmentation:",
|
877 |
-
("Upload Scan", "Example Scan"),
|
878 |
-
)
|
879 |
-
import pyvista as pv
|
880 |
-
if inputs == "Example Scan":
|
881 |
-
st.markdown("Expected time per prediction: 7-10 min.")
|
882 |
-
mesh = pv.read("ZOUIF2W4_upper.obj")
|
883 |
-
plotter = pv.Plotter()
|
884 |
-
|
885 |
-
# Add the mesh to the plotter
|
886 |
-
plotter.add_mesh(mesh, color='white', show_edges=False)
|
887 |
-
segment = st.button(
|
888 |
-
"✔️ Submit",
|
889 |
-
help="Submit 3D scan for segmentation",
|
890 |
-
)
|
891 |
-
with st.expander("View Scan", expanded=False):
|
892 |
-
stpyvista(plotter)
|
893 |
-
|
894 |
-
if segment:
|
895 |
-
segmentation_main("ZOUIF2W4_upper.obj")
|
896 |
-
|
897 |
-
|
898 |
-
|
899 |
-
elif inputs == "Upload Scan":
|
900 |
-
file = st.file_uploader("Please upload an OBJ Object file", type=["OBJ"])
|
901 |
-
st.markdown("Expected time per prediction: 7-10 min.")
|
902 |
-
if file is not None:
|
903 |
-
# save the uploaded file to disk
|
904 |
-
with open("file.obj", "wb") as buffer:
|
905 |
-
shutil.copyfileobj(file, buffer)
|
906 |
-
# 复制数据
|
907 |
-
obj_path = "file.obj"
|
908 |
-
|
909 |
-
mesh = pv.read(obj_path)
|
910 |
-
plotter = pv.Plotter()
|
911 |
-
|
912 |
-
# Add the mesh to the plotter
|
913 |
-
plotter.add_mesh(mesh, color='white', show_edges=False)
|
914 |
-
segment = st.button(
|
915 |
-
"✔️ Submit",
|
916 |
-
help="Submit 3D scan for segmentation",
|
917 |
-
)
|
918 |
-
with st.expander("View Scan", expanded=False):
|
919 |
-
stpyvista(plotter)
|
920 |
-
|
921 |
-
if segment:
|
922 |
-
segmentation_main(obj_path)
|
923 |
-
|
924 |
-
|
925 |
-
|
926 |
-
|
927 |
|
928 |
if __name__ == "__main__":
|
929 |
-
app =
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import streamlit as st
|
2 |
+
from streamlit import session_state as session
|
3 |
|
4 |
from PIL import Image
|
5 |
|
|
|
8 |
# Font
|
9 |
with open("utils/style.css") as css:
|
10 |
st.markdown(f"<style>{css.read()}</style>", unsafe_allow_html=True)
|
11 |
+
|
12 |
# Logo
|
13 |
self.image_path = "utils/teeth-295404_1280.png"
|
14 |
self.image = Image.open(self.image_path)
|
|
|
30 |
unsafe_allow_html=True,
|
31 |
)
|
32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
# Configure Streamlit page
|
34 |
+
st.set_page_config(page_title="Teeth Segmentation", page_icon="ⓘ")
|
35 |
|
36 |
+
class Intro(TeethApp):
|
37 |
def __init__(self):
|
38 |
TeethApp.__init__(self)
|
39 |
self.build_app()
|
40 |
|
41 |
def build_app(self):
|
42 |
+
st.title("AI-assited Tooth Segmentation")
|
43 |
+
st.markdown("This app automatically segments intra-oral scans of teeth using machine learning.")
|
44 |
+
st.markdown("Head to the 'Segment' tab to try it out!")
|
45 |
+
st.markdown("**Example:**")
|
46 |
+
st.image("illu.png")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
if __name__ == "__main__":
|
49 |
+
app = Intro()
|