当前位置:网站首页>Open3d learning note 3 [sampling and voxelization]
Open3d learning note 3 [sampling and voxelization]
2022-07-02 07:54:00 【Silent clouds】
open3d Voxelization of learning notes
One 、 Add some small knowledge
1、 With mesh Mode reading ply file
import open3d as o3d
mesh = o3d.io.read_triangle_mesh("mode/Fantasy Dragon.ply")
mesh.compute_vertex_normals()
2. Rotation matrix
The 3D model uses R,T Two parameters to transform , The spatial coordinate system of the view is established : Up for z Axis , To the right is y Axis ,x The axis points to the front of the screen . Use transform
Method transform coordinates , The transformation matrix is [4*4] Matrix ,transform([[R, T], [0, 1]])
.
Read one normally ply file :
import open3d as o3d
pcd = o3d.io.read_point_cloud("mode/Fantasy Dragon.ply")
o3d.visualization.draw_geometries([pcd], width=1280, height=720)
The display effect is as shown in the figure :
Use the conversion function , Put him horizontally , And the head faces the screen . Then it is to change the original z Shaft change to y Axis ,y Shaft change to x Axis ,x Shaft change to z Axis , So the code is :
import open3d as o3d
mesh = o3d.io.read_triangle_mesh("mode/Fantasy Dragon.ply")
mesh.compute_vertex_normals()
mesh.transform([[0, 1, 0, 0], [0, 0, 1, 0], [1, 0, 0, 0], [0, 0, 0, 1]])
o3d.visualization.draw_geometries([mesh], width=1280, height=720)
effect :
Two 、 The way to convert to point cloud
1、 Turn into numpy The array is redrawn into a point cloud
import open3d as o3d
import numpy as np
mesh = o3d.io.read_triangle_mesh("mode/Fantasy Dragon.ply")
mesh.compute_vertex_normals()
v_mesh = np.asarray(mesh.vertices)
pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(v_mesh)
o3d.visualization.draw_geometries([pcd], width=1280, height=720)
about ply The format of the file is ok , But if stl This triangular mesh , The result of transformation will be a little unsatisfactory .
2、 sampling
open3d Provides a sampling method , Sampling points can be set , simplified model .
import open3d as o3d
mesh = o3d.io.read_triangle_mesh("mode/ganyu.STL")
mesh.compute_vertex_normals()
pcd = o3d.geometry.TriangleMesh.sample_points_uniformly(mesh, number_of_points=10000) # Sampling point cloud
o3d.visualization.draw_geometries([pcd], width=1280, height=720)
3、 ... and 、 Voxelization
Voxelization , Can simplify the model , Get a uniform mesh .
Convert triangle mesh to voxel mesh
import open3d as o3d
import numpy as np
print("Load a ply point cloud, print it, and render it")
mesh = o3d.io.read_triangle_mesh("mode/ganyu.STL")
mesh.compute_vertex_normals()
mesh.scale(1 / np.max(mesh.get_max_bound() - mesh.get_min_bound()), center=mesh.get_center())
voxel_grid = o3d.geometry.VoxelGrid.create_from_triangle_mesh(mesh, voxel_size=0.05)
o3d.visualization.draw_geometries([voxel_grid], width=1280, height=720)
Point cloud generates voxel mesh
import open3d as o3d
import numpy as np
print("Load a ply point cloud, print it, and render it")
pcd = o3d.io.read_point_cloud("mode/Fantasy Dragon.ply")
pcd.scale(1 / np.max(pcd.get_max_bound() - pcd.get_min_bound()), center=pcd.get_center())
pcd.colors = o3d.utility.Vector3dVector(np.random.uniform(0,1,size=(2000,3)))
print('voxelization')
voxel_grid = o3d.geometry.VoxelGrid.create_from_point_cloud(pcd, voxel_size=0.05)
o3d.visualization.draw_geometries([voxel_grid], width=1280, height=720)
Four 、 Vertex normal estimation
voxel_down_pcd = pcd.voxel_down_sample(voxel_size=0.05)
voxel_down_pcd.estimate_normals(
search_param=o3d.geometry.KDTreeSearchParamHybrid(radius=0.1, max_nn=30))
o3d.visualization.draw_geometries([voxel_down_pcd], point_show_normal=True, width=1280, height=720)
estimate_normals
Calculate the normal of each point . This function finds adjacent points and calculates the principal axis of adjacent points using covariance analysis .
This function will KDTreeSearchParamHybrid
Class as a parameter . The two key parameters are the specified search radius and the maximum nearest neighbor .radius=0.1, max_nn=30
That is to 10cm Search radius for , And only consider 30 Adjacent points to save computing time .
Read the normal vector
print(" Print the first vector :")
print(voxel_down_pcd.normals[0])
# Print the first vector :
#[ 0.51941952 0.82116269 -0.23642166]
# Print the first ten normal vectors
print(np.asarray(voxel_down_pcd.normals)[:10,:])
5、 ... and 、 What should be noted
- pcd The format file belongs to the point cloud type ,ply It can be read in point cloud and grid mode at the same time , use mesh When reading , It can be treated as triangular mesh , use pcd Reading can be directly used as point cloud data processing .
- When triangle meshes are directly sampled and then normals are calculated, there will be errors in normal annotation , That is, all normals point in one direction .
- To avoid sampling normal errors, use
sample_points_poisson_disk()
Method sampling .
边栏推荐
- Yolov3 trains its own data set (mmdetection)
- How gensim freezes some word vectors for incremental training
- 【Wing Loss】《Wing Loss for Robust Facial Landmark Localisation with Convolutional Neural Networks》
- 【Sparse-to-Dense】《Sparse-to-Dense:Depth Prediction from Sparse Depth Samples and a Single Image》
- Implementation of yolov5 single image detection based on pytorch
- 【双目视觉】双目矫正
- Replace convolution with full connection layer -- repmlp
- CPU的寄存器
- Replace self attention with MLP
- latex公式正体和斜体
猜你喜欢
Faster-ILOD、maskrcnn_ Benchmark installation process and problems encountered
Network metering - transport layer
Timeout docking video generation
【Cutout】《Improved Regularization of Convolutional Neural Networks with Cutout》
【DIoU】《Distance-IoU Loss:Faster and Better Learning for Bounding Box Regression》
[in depth learning series (8)]: principles of transform and actual combat
Open3D学习笔记一【初窥门径,文件读取】
open3d学习笔记三【采样与体素化】
用全连接层替代掉卷积 -- RepMLP
Faster-ILOD、maskrcnn_benchmark安装过程及遇到问题
随机推荐
conda常用命令
[learning notes] matlab self compiled image convolution function
What if the laptop task manager is gray and unavailable
【MagNet】《Progressive Semantic Segmentation》
Comparison of chat Chinese corpus (attach links to various resources)
Win10 solves the problem that Internet Explorer cannot be installed
Jetson nano installation tensorflow stepping pit record (scipy1.4.1)
【Sparse-to-Dense】《Sparse-to-Dense:Depth Prediction from Sparse Depth Samples and a Single Image》
MMDetection模型微调
[multimodal] clip model
【Cutout】《Improved Regularization of Convolutional Neural Networks with Cutout》
win10解决IE浏览器安装不上的问题
Replace self attention with MLP
程序的内存模型
CONDA common commands
【MobileNet V3】《Searching for MobileNetV3》
【BiSeNet】《BiSeNet:Bilateral Segmentation Network for Real-time Semantic Segmentation》
【Paper Reading】
What if the notebook computer cannot run the CMD command
Thesis tips