CN116309144A - Point cloud shape completion method based on diffusion probability model - Google Patents

Point cloud shape completion method based on diffusion probability model Download PDF

Info

Publication number
CN116309144A
CN116309144A CN202310222852.4A CN202310222852A CN116309144A CN 116309144 A CN116309144 A CN 116309144A CN 202310222852 A CN202310222852 A CN 202310222852A CN 116309144 A CN116309144 A CN 116309144A
Authority
CN
China
Prior art keywords
point cloud
point
noise
diffusion
points
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310222852.4A
Other languages
Chinese (zh)
Inventor
唐可可
何旭
钟承志
彭伟龙
李树栋
李默涵
王乐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangzhou University
Original Assignee
Guangzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangzhou University filed Critical Guangzhou University
Priority to CN202310222852.4A priority Critical patent/CN116309144A/en
Publication of CN116309144A publication Critical patent/CN116309144A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00—Image enhancement or restoration
    • G06T5/70—Denoising; Smoothing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/84—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using probabilistic graphical models from image or video features, e.g. Markov models or Bayesian networks
    • G06V10/85—Markov-related models; Markov random fields
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Image Processing (AREA)

Abstract

The invention relates to the field of robot vision, and discloses a point cloud shape complement method based on a diffusion probability model, which comprises the following steps: s1: randomly deleting a certain proportion of points from the whole point cloud in the data set to generate a incomplete point cloud p for training; s2: inputting the residual point cloud into a trained encoder to obtain a feature vector z corresponding to the residual point cloud; s3: inputting the residual point cloud into a neural network, and generating a noise point cloud according to parameters mu and sigma of the input initialization noise point cloud; s4: designing a loss function, and training a Markov chain model; s5: and gradually reducing the noise of each point in the noise point cloud according to the Markov chain model, and generating a completed point cloud when the output of the neural network is not obviously changed. According to the invention, the characteristics in the residual point cloud are extracted to carry out back diffusion on the noise point cloud, so that the completion work of the point cloud is completed with higher precision and reduction degree.

Description

Point cloud shape completion method based on diffusion probability model
Technical Field
The invention relates to the field of robot vision, in particular to a point cloud shape complement method based on a diffusion probability model.
Background
The point cloud is a method for representing the shape of the surface of an object by computer technology, and a three-dimensional coordinate system is generally used, the surface of the object is scattered into a plurality of points, and each point is marked with coordinates, so that the three-dimensional shape of the object can be constructed by a computer. The point cloud technology is widely applied to the fields of three-dimensional modeling, robot navigation and the like, and can be used for quickly establishing a three-dimensional model and measuring and analyzing an object. In the data acquisition process, the 3D laser scanner is affected by the characteristics of the measured object, the measurement method and the environment, the point cloud is inevitably lost, the stability of the 3D scanner in the scanning process also has special influence on the scanning point cloud, and the continuous rotation of the scaffold, the mechanical structure and the scanner inevitably causes mechanical shake, which can affect the echo and the deviation between the position of the acquired point cloud and the actual measured object. However, many tasks such as object classification and robot grabbing depend on the complete 3D shape of the object, so the 3D point cloud complement technology is a very basic and important content in 3D perception, and the point cloud complement refers to a technology of completing missing parts in a point cloud model through computer technology, and aims to predict and recover the complete 3D shape based on the scanned incomplete point cloud segments.
The existing point cloud complement technical schemes mainly comprise 4 kinds: the prior technical proposal mainly comprises 4 kinds:
1. fitting-based methods: generating a complete point cloud model by carrying out curved surface or polyhedron fitting on the point cloud; 2. interpolation-based methods: filling a vacant area in the point cloud by carrying out interpolation operation on the point cloud data, and generating a complete point cloud model; 3. learning-based methods: training the point cloud data by using a machine learning algorithm, and learning the characteristics of the point cloud to generate a complete point cloud model; 4. optimization-based methods: and (3) solving an optimal point cloud completion result by defining an objective function of point cloud completion and using an optimization algorithm.
Under the conditions of high scene complexity and uneven noise and sampling rate, the fitting-based method may have the problems of inaccurate fitting, excessively simplified or complex fitting result and the like; under the conditions of data noise and uneven sampling rate, the interpolation-based method may have the problems of discontinuous interpolation results, excessive smoothness or insufficient smoothness and the like; under the conditions of low quality of training data, unbalanced data distribution and unsuitable network structure, the problems of over fitting, under fitting, insufficient training data quantity and the like possibly occur in a learning-based method; because the optimization problem generally has the difficult properties of high dimensionality and non-convexity and the special properties of the point cloud data, the optimization-based method can have the problems of strong dependence on the initial point cloud, high calculation complexity, difficult parameter adjustment and the like.
Disclosure of Invention
Aiming at the defects of the prior point cloud completion method in the background art, the invention provides a point cloud shape completion method based on a diffusion probability model, which solves the problems of inaccurate fitting, too simple or complex fitting result, insufficient training data amount, strong dependence on initial point cloud, high calculation complexity and the like in the prior art, and can realize high-precision and more detailed reduction on the generated point cloud, thereby ensuring the quality of subsequent point cloud reconstruction, three-dimensional model reconstruction, local space information extraction and subsequent processing.
The invention provides the following technical scheme: s1: randomly deleting a certain proportion of points from the whole point cloud in the data set to generate a incomplete point cloud p for training; s2: inputting the residual point cloud into a trained encoder to obtain a feature vector z corresponding to the residual point cloud; s3: inputting the residual point cloud into a neural network, and generating a noise point cloud according to the input initialized parameters mu and sigma of the noise point cloud; wherein, the noise point cloud obeys Gaussian distribution, mu represents the mean value of the distribution, and sigma represents the standard deviation of the distribution; s4: designing a loss function, and training a Markov chain model; s5: and gradually reducing the noise of each point in the noise point cloud according to the Markov chain model, and generating a completed point cloud when the output of the neural network is not obviously changed.
Preferably, the data sets are ModelNet40 and ShapeNetPart.
Preferably, the method for supplementing the shape of the point cloud based on the diffusion probability model, S2 includes:
inputting the incomplete point cloud into a trained encoder phi, wherein the encoder phi maps the incomplete point cloud from a high-dimensional sample space to a low-dimensional feature space, and one incomplete point cloud correspondingly generates a one-dimensional feature vector z; the mathematical expression of the encoder phi extraction features is as follows:
z＝φ(p)，
wherein z is a one-dimensional feature vector, phi is an encoder, and p is a fault point cloud.
Preferably, the method for supplementing the shape of the point cloud based on the diffusion probability model, S3 includes:
random sampling is carried out from the Gaussian distribution point cloud to generate a random initial noise point cloud, the neural network delta is used for learning how to obtain parameters mu and sigma of the noise point cloud through the residual point cloud,
Figure BDA0004121886650000031
wherein μ is a mean value of the distribution, σ is a standard deviation of the distribution, δ is a neural network, and p is a fault point cloud.
Preferably, the method for supplementing the shape of the point cloud based on the diffusion probability model, S4 includes:
the chamfer distance loss (Chamfer Distance Loss) and the hausdorff distance loss (Hausdorff Distance Loss) are used as a loss function of model training to limit the distance between the original full point cloud and the post-completion point cloud.
Preferably, the process of gradually generating the completed complete point cloud according to the characteristics of the obtained residual point cloud and the noise point cloud is a back diffusion process, and the Markov chain model is the back diffusion process;
preferably, the mathematical expression of the chamfer distance loss is as follows:
Figure BDA0004121886650000032
wherein X and Y are two point clouds, X is any point in point cloud X, Y is any point in point cloud Y,
Figure BDA0004121886650000041
representing the distance of point x to the nearest point in the set of points Y, +.>
Figure BDA0004121886650000042
Representing the distance of the representative point y to the nearest point in the set of points X;
the mathematical expression of the Haoskov distance is as follows:
Figure BDA0004121886650000043
wherein X and Y are two point clouds, X and Y are any point of the point clouds X and Y respectively,
Figure BDA0004121886650000044
representing the distance of point x to the nearest point in the set of points Y, +.>
Figure BDA0004121886650000045
Represents the maximum value of the distances from all points in the point set X to the nearest point in the point set Y. Likewise, a->
Figure BDA0004121886650000046
Representing the distance of point y to the nearest point in point set X, +.>
Figure BDA0004121886650000047
Represents the maximum value of the distances from all points in the set of points Y to the nearest point in the set of points X.
Preferably, the method for supplementing the shape of the point cloud based on the diffusion probability model, S5 includes:
gradually generating a completed complete point cloud according to the characteristics of the obtained residual point cloud and the noise point cloud, wherein the process is a back diffusion process, the whole process is represented by a Markov chain model, and each point x in the noise point cloud is subjected to noise reduction by the following transition probability:
Figure BDA0004121886650000048
wherein t represents the number of steps of back diffusion, xi represents each point in the point cloud, and x i (t) And (3) representing the point cloud of the t step of back diffusion, wherein θ is a parameter learned from a neural network, and z is a feature vector of the incomplete point cloud.
The invention has the following beneficial effects:
1. the method for complementing the residual point cloud through the diffusion probability model is provided, and the noise point cloud is subjected to inverse diffusion through extracting the characteristics in the residual point cloud, so that the complementing work of the point cloud is completed with higher precision and reduction degree.
2. Compared with the existing shape complement work for three-point cloud, the point cloud generated by the complement method has high precision and restores more details, and further can ensure the quality of subsequent point cloud reconstruction, three-dimensional model reconstruction, local space information extraction and subsequent processing.
Drawings
Fig. 1 is a working frame diagram of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The point cloud shape complement method based on the diffusion probability model is characterized in that the point cloud is regarded as particles in a thermodynamic system in contact with a thermal reservoir under the inspired by a diffusion process in unbalanced thermodynamics, and the particles diffuse from original distribution to noise distribution, so that the generation and complement process of the point cloud is equivalent to a back diffusion process for converting the noise distribution into required distribution. Modeling of this inverse process is a Markov chain conditioned on some hidden shape. When noise is introduced into the point cloud, the point cloud gradually changes from complex distribution related to shape to Gaussian distribution noise, and the process is called forward diffusion. The shape complement model of the point cloud can be regarded as a reverse diffusion process, partial shape features in the learning residual point cloud are extracted through a trained deep neural network, and the extracted features are used for gradually reducing noise of a noise point cloud sampled from Gaussian distribution until the point cloud recovers the complete shape.
Referring to fig. 1, the present invention provides the following technical solutions: a point cloud shape complement method based on a diffusion probability model comprises the following steps:
s1: randomly deleting a certain proportion (for example, 40%) of points from the whole point cloud in the data set to generate a incomplete point cloud p for training; the data sets are ModelNet40 and ShapeNetPart;
s2: inputting the residual point cloud into a trained encoder to obtain a feature vector z corresponding to the residual point cloud;
specifically, the point cloud shape complement method based on the diffusion probability model, S2 includes:
inputting the incomplete point cloud into a trained encoder phi, wherein the encoder phi maps the incomplete point cloud from a high-dimensional sample space to a low-dimensional feature space, and one incomplete point cloud correspondingly generates a one-dimensional feature vector z; the mathematical expression of the encoder phi extraction features is as follows:
z＝φ(p)，
wherein z is a one-dimensional feature vector, phi is an encoder, and p is a fault point cloud.
S3: inputting the residual point cloud into a neural network, and generating a noise point cloud according to the input initialized parameters mu and sigma of the noise point cloud; wherein, the noise point cloud obeys Gaussian distribution, mu represents the mean value of the distribution, and sigma represents the standard deviation of the distribution;
specifically, random sampling is performed from the Gaussian distribution point cloud to generate a random initial noise point cloud, the parameters mu and sigma of the noise point cloud are learned through the neural network delta, the parameters mu and sigma of the noise point cloud are obtained through the residual point cloud, and the mathematical expression of the learning process is as follows:
μ,σ＝δ(p)，
wherein μ is a mean value of the distribution, σ is a standard deviation of the distribution, δ is a neural network, and p is a fault point cloud.
S4: designing a loss function, and training a Markov chain model;
specifically, model training mainly uses chamfer distance loss (Chamfer Distance Loss) and hausdorff distance loss (Hausdorff Distance Loss) as training loss functions, mainly to limit the distance between the original complete point cloud and the complement back point cloud;
the method comprises the steps of obtaining a residual point cloud, wherein the process of gradually generating a completed complete point cloud according to the characteristics of the obtained residual point cloud and the noise point cloud is a back diffusion process, and the Markov chain model is the back diffusion process;
specifically, the mathematical expression of the chamfer distance loss is as follows:
Figure BDA0004121886650000071
wherein X and Y are two point clouds, X is any point of point cloud X, Y is any point of point cloud Y,
Figure BDA0004121886650000072
representing the distance of point x to the nearest point in the set of points Y, +.>
Figure BDA0004121886650000073
Representing the distance of the point y to the nearest point in the set of points X.
The mathematical expression of the Haoskov distance is as follows:
Figure BDA0004121886650000074
wherein X and Y are two point clouds, X and Y are any point of the point clouds X and Y respectively,
Figure BDA0004121886650000075
representing the distance of point x to the nearest point in the set of points Y, +.>
Figure BDA0004121886650000076
Represents the maximum value of the distances from all points in the point set X to the nearest point in the point set Y. Likewise, a->
Figure BDA0004121886650000077
Representing the distance of point y to the nearest point in point set X, +.>
Figure BDA0004121886650000078
Represents the maximum value of the distances from all points in the set of points Y to the nearest point in the set of points X.
S5: gradually reducing noise of each point in the noise point cloud according to the Markov chain model, and generating a complete point cloud after a sufficient number of steps; when the Hausdorff distance between the point cloud generated in the current round and the point cloud generated in the previous round is smaller than 0.0001, the output of the neural network can be considered to be no longer obviously changed, and the complete point cloud can be generated;
specifically, each point x in the noise point cloud is denoised by the following transition probabilities:
Figure BDA0004121886650000079
wherein t represents the number of steps of back diffusion, xi represents each point in the point cloud, and x i (t) And (3) representing the point cloud of the t step of back diffusion, wherein θ is a parameter learned from a neural network, and z is a feature vector of the incomplete point cloud.
The invention discloses a point cloud shape complement method based on a diffusion probability model, which is inspired by a diffusion process in unbalanced thermodynamics, models the whole model based on a Markov chain, carries out forward diffusion by adding noise to an original point cloud, and trains a network model. After training is completed, inputting the residual point cloud into a trained encoder and a neural network to obtain feature vectors of the point cloud and initial parameters of the noise point cloud, and then generating the completed point cloud through a back diffusion process. According to the invention, incomplete point cloud completion is performed through the diffusion probability model, namely, how to perform point cloud completion is learned through forward diffusion by extracting the characteristics of the point cloud through the encoder, and noise point cloud is restored to complete point cloud through reverse diffusion, so that high precision of generated point cloud is ensured, and more details are restored.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims (8)

1. A point cloud shape complement method based on a diffusion probability model comprises the following steps:
s1: randomly deleting a certain proportion of points from the whole point cloud in the data set to generate a incomplete point cloud p for training;
s2: inputting the residual point cloud into a trained encoder to obtain a feature vector z corresponding to the residual point cloud;
s3: inputting the residual point cloud into a neural network, and generating a noise point cloud according to the input initialized parameters mu and sigma of the noise point cloud; wherein, the noise point cloud obeys Gaussian distribution, mu represents the mean value of the distribution, and sigma represents the standard deviation of the distribution;
s4: designing a loss function, and training a Markov chain model;
s5: and gradually reducing the noise of each point in the noise point cloud according to the Markov chain model, and generating a completed point cloud when the output of the neural network is not obviously changed.
2. The point cloud shape completion method based on the diffusion probability model according to claim 1, wherein: the data sets are ModelNet40 and ShapeNetPart.
3. The point cloud shape completion method based on the diffusion probability model according to claim 1, wherein S2 comprises:
inputting the incomplete point cloud into a trained encoder phi, wherein the encoder phi maps the incomplete point cloud from a high-dimensional sample space to a low-dimensional feature space, and one incomplete point cloud correspondingly generates a one-dimensional feature vector z; the mathematical expression of the encoder phi extraction features is as follows:
z＝φ(p)，
wherein z is a one-dimensional feature vector, phi is an encoder, and p is a fault point cloud.
4. The point cloud shape completion method based on the diffusion probability model of claim 1, wherein S3 comprises:
random sampling is carried out from the Gaussian distribution point cloud to generate a random initial noise point cloud, the neural network delta is used for learning how to obtain parameters mu and sigma of the noise point cloud through the residual point cloud,
the mathematical expression of the learning process is as follows:
μ,σ＝δ(p)，
wherein μ is a mean value of the distribution, σ is a standard deviation of the distribution, δ is a neural network, and p is a fault point cloud.
5. The point cloud shape completion method based on the diffusion probability model according to claim 1, wherein S4 comprises: and taking the chamfer distance loss and the Hausdorff distance loss as loss functions of model training to limit the distance between the original complete point cloud and the complementary point cloud.
6. The point cloud shape completion method based on the diffusion probability model according to claim 1, wherein: the process of gradually generating the completed complete point cloud according to the characteristics of the obtained residual point cloud and the noise point cloud is a back diffusion process, and the Markov chain model is the back diffusion process.
7. The loss function of claim 5, wherein the mathematical expression of the chamfer distance loss is as follows:
Figure FDA0004121886640000021
wherein X and Y are two point clouds, X is any point of point cloud X, Y is any point of point cloud Y,
Figure FDA0004121886640000022
representing the distance of point x to the nearest point in the set of points Y, +.>
Figure FDA0004121886640000023
Representing the distance of the representative point y to the nearest point in the set of points X;
the mathematical expression of the Haoskov distance loss is as follows:
Figure FDA0004121886640000024
wherein X and Y are two point clouds, X and Y are any point of the point clouds X and Y respectively,
Figure FDA0004121886640000025
representing the distance of point x to the nearest point in the set of points Y, +.>
Figure FDA0004121886640000026
Represents the maximum value of the distances from all points in the point set X to the nearest point in the point set Y. Likewise, a->
Figure FDA0004121886640000027
Representing the distance of point y to the nearest point in point set X, +.>
Figure FDA0004121886640000031
Represents the maximum value of the distances from all points in the set of points Y to the nearest point in the set of points X.
8. The point cloud shape completion method based on the diffusion probability model according to claim 1, wherein S5 comprises:
each point x in the noise point cloud is denoised by the following transition probabilities:
Figure FDA0004121886640000032
wherein t represents the number of steps of back diffusion, xi represents each point in the point cloud, and x i (t) And (3) representing the point cloud of the t step of back diffusion, wherein θ is a parameter learned from a neural network, and z is a feature vector of the incomplete point cloud.
CN202310222852.4A 2023-03-08 2023-03-08 Point cloud shape completion method based on diffusion probability model Pending CN116309144A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310222852.4A CN116309144A (en) 2023-03-08 2023-03-08 Point cloud shape completion method based on diffusion probability model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310222852.4A CN116309144A (en) 2023-03-08 2023-03-08 Point cloud shape completion method based on diffusion probability model

Publications (1)

Publication Number Publication Date
CN116309144A true CN116309144A (en) 2023-06-23

Family

ID=86802571

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310222852.4A Pending CN116309144A (en) 2023-03-08 2023-03-08 Point cloud shape completion method based on diffusion probability model

Country Status (1)

Country Link
CN (1) CN116309144A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117058464A (en) * 2023-08-31 2023-11-14 强联智创（北京）科技有限公司 Method and device for training generation model for generating healthy blood vessel surface
CN117789198A (en) * 2024-02-28 2024-03-29 上海几何伙伴智能驾驶有限公司 Method for realizing point cloud degradation detection based on 4D millimeter wave imaging radar
CN117975202A (en) * 2024-04-01 2024-05-03 之江实验室 Model training method, service execution method, device, medium and equipment

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117058464A (en) * 2023-08-31 2023-11-14 强联智创（北京）科技有限公司 Method and device for training generation model for generating healthy blood vessel surface
CN117058464B (en) * 2023-08-31 2024-06-11 强联智创（北京）科技有限公司 Method and device for training generation model for generating healthy blood vessel surface
CN117789198A (en) * 2024-02-28 2024-03-29 上海几何伙伴智能驾驶有限公司 Method for realizing point cloud degradation detection based on 4D millimeter wave imaging radar
CN117789198B (en) * 2024-02-28 2024-05-14 上海几何伙伴智能驾驶有限公司 Method for realizing point cloud degradation detection based on 4D millimeter wave imaging radar
CN117975202A (en) * 2024-04-01 2024-05-03 之江实验室 Model training method, service execution method, device, medium and equipment

Similar Documents

Publication Publication Date Title
CN116309144A (en) Point cloud shape completion method based on diffusion probability model
CN112241997B (en) Three-dimensional model building and repairing method and system based on multi-scale point cloud up-sampling
CN108399649B (en) Single-picture three-dimensional face reconstruction method based on cascade regression network
CN109903327B (en) Target size measurement method of sparse point cloud
CA2423212C (en) Apparatus and method for generating a three-dimensional representation from a two-dimensional image
CN110060329B (en) Mobile terminal human body model reconstruction method based on color depth video stream data
CN111553858B (en) Image restoration method and system based on generation countermeasure network and application thereof
CN112613097A (en) BIM rapid modeling method based on computer vision
WO2018156126A1 (en) Real-time generation of synthetic data from multi-shot structured light sensors for three-dimensional object pose estimation
CN110634161A (en) Method and device for quickly and accurately estimating pose of workpiece based on point cloud data
CN111612888B (en) Automatic generation method, system and storage medium for cultural relic building graph
CN114842136A (en) Single-image three-dimensional face reconstruction method based on differentiable renderer
Tabib et al. Learning-based hole detection in 3D point cloud towards hole filling
CN112329726A (en) Face recognition method and device
CN110047146B (en) Error correction method based on single revolving body image 3D restoration
CN117475110B (en) Semantic three-dimensional reconstruction method and device for blade, electronic equipment and storage medium
CN111369670A (en) Method for real-time construction of practical training digital twin model
CN112562067A (en) Method for generating large-batch point cloud data sets
CN116740665A (en) Point cloud target detection method and device based on three-dimensional cross-correlation ratio
CN115797414A (en) Complex curved surface measurement point cloud data registration method considering measuring head radius
Yang et al. Center extraction algorithm of linear structured light stripe based on improved gray barycenter method
CN114882095A (en) Object height online measurement method based on contour matching
CN114494576A (en) Rapid high-precision multi-view face three-dimensional reconstruction method based on implicit function
Cui et al. Hole Inpainting Algorithm for Half-Organized Point Cloud Obtained by Structured-Light Section System
CN110689618A (en) Three-dimensional deformable object filling method based on multi-scale variational graph convolution

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination