WO2024113169A1 - Few-shot point cloud semantic segmentation method, and network, storage medium and processor - Google Patents

Few-shot point cloud semantic segmentation method, and network, storage medium and processor Download PDF

Info

Publication number
WO2024113169A1
WO2024113169A1 PCT/CN2022/135087 CN2022135087W WO2024113169A1 WO 2024113169 A1 WO2024113169 A1 WO 2024113169A1 CN 2022135087 W CN2022135087 W CN 2022135087W WO 2024113169 A1 WO2024113169 A1 WO 2024113169A1
Authority
WO
WIPO (PCT)
Prior art keywords
prototype
feature
features
query
calibration
Prior art date
Application number
PCT/CN2022/135087
Other languages
French (fr)
Chinese (zh)
Inventor
王体超
郝富生
程俊
张锲石
吴福祥
Original Assignee
中国科学院深圳先进技术研究院
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 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Priority to PCT/CN2022/135087 priority Critical patent/WO2024113169A1/en
Publication of WO2024113169A1 publication Critical patent/WO2024113169A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T15/00—3D [Three Dimensional] image rendering
    • G06T15/10—Geometric effects
    • G06T15/30—Clipping

Definitions

  • the present application belongs to the technical field of point cloud semantic segmentation, and in particular, relates to a small sample point cloud semantic segmentation method, network, storage medium and processor.
  • the current small sample point cloud semantic segmentation technology mainly uses the prototype features of the annotated support set data to propagate labels on the query set data to obtain the corresponding point cloud labels.
  • the current small sample point cloud semantic segmentation technology mainly uses the prototype features of the annotated support set data to propagate labels on the query set data to obtain the corresponding point cloud labels.
  • the purpose of this application is to provide a small sample point cloud semantic segmentation method, network and storage medium and processor, which can flexibly cooperate with the multiple functions assigned by secondary users and regulate the corresponding system resources to derive the reachable rate area for the information security scenario of the cognitive system and solve the technical problem of establishing a secure coexistence relationship between authorized users and cognitive users.
  • the present application provides a small sample point cloud semantic segmentation method, comprising the following steps:
  • the distance between the calibration query feature and the calibration prototype feature is measured by a label propagation module, and label propagation is performed to achieve semantic segmentation of the query set point cloud.
  • the present application also provides a small sample point cloud semantic segmentation network using the above semantic segmentation method, including a sequentially connected feature extractor, a prototype amplification module, a feature calibration module and a label propagation module;
  • the feature extractor extracts features from the support set point cloud and the query set point cloud, and outputs corresponding support set features and query set features;
  • the prototype augmentation module combines the support set features and the query set features to obtain augmented multiple prototypes
  • the feature calibration module exchanges information between the multiple prototypes and the query set features to obtain calibration prototype features and calibration query features
  • the present application also provides a storage medium, which stores a program file that can implement the above-mentioned small sample point cloud semantic segmentation method.
  • the present application proposes a small sample point cloud semantic segmentation method, network, storage medium and processor.
  • the present application uses label propagation to extract pseudo-prototype features that are suitable for query set data, thereby obtaining prototype features that are suitable for query set data, and performs feature calibration by extracting the relationship between the prototype and the query set data.
  • Prototype expansion effectively utilizes the distribution information of the query set data and the prototype information of the support set.
  • the adaptability of the prototype to the query set data is further improved. Therefore, the present invention can obtain prototype features that are suitable for the query set and realize effective segmentation of the point cloud scene.
  • FIG1 is a schematic diagram of the process of the semantic segmentation method of a small sample point cloud of the present application
  • FIG2 is a schematic diagram of the architecture of the prototype amplification module of the present application.
  • FIG3 is a schematic diagram of the architecture of the feature calibration module of the present application.
  • FIG4 is a schematic diagram of the main steps of the small sample point cloud semantic segmentation method of this application.
  • Embodiment 1 is a diagrammatic representation of Embodiment 1:
  • FIG1-4 shows the implementation process of the small sample point cloud semantic segmentation method provided in the first embodiment of the present application. For the sake of convenience, only the part related to the embodiment of the present application is shown, which is described in detail as follows:
  • the present application provides a small sample point cloud semantic segmentation method, comprising the following steps:
  • the amplified multi-prototype and the query set features are input into a feature calibration module to obtain calibration prototype features and calibration query features;
  • the distance between the calibration query feature and the calibration prototype feature is measured by a label propagation module, and label propagation is performed to achieve semantic segmentation of the query set point cloud.
  • FIG1 the solution flow of the present invention is shown in FIG1 .
  • the support set data and the query set data are passed through a shared feature extractor to obtain the corresponding support set features f s and query set features f q .
  • the expanded multiple prototypes p and query set features f q are input into the feature calibration module to obtain the calibration prototype features and calibration query features
  • semantic segmentation of point cloud is achieved through label propagation.
  • step s2 comprises the following steps:
  • the support set features are aggregated with the support set mask to obtain multiple prototypes of the support set;
  • the query set features are aggregated with the pseudo labels of the query set to obtain a pseudo prototype
  • the structure of the prototype augmentation module is shown in Figure 2.
  • the query set features are aggregated in the same way to obtain the pseudo-prototype.
  • the feature aggregation includes the following steps:
  • the feature calibration module realizes the information flow between the multi-prototype information and the query set features through two shared residual cross-attention modules, and then outputs the calibrated prototype features and the calibrated query features.
  • the residual cross attention module obtains attention features and query vector features through three convolution kernels, and uses the attention mechanism to extract the relationship matrix between input features, and uses the query vector to extract channel information.
  • Figure 3(a) is the specific architecture of the feature calibration module, which realizes the information flow between the prototype information and the query set features through two shared residual attention modules;
  • Figure 3(b) is the specific structure of the residual cross attention module, which obtains the attention features and query vector features through three convolution kernels.
  • the feature calibration module contains two shared residual cross attention modules to achieve information exchange between the prototype feature p and the query set feature fq .
  • the residual cross attention module uses the attention mechanism to extract the relationship matrix between the input features and uses the query vector to extract channel information. Taking prototype calibration as an example, the calculation method is as follows:
  • Q p k 1 (p)
  • K p k 2 (f q )
  • V p k 1 (p)
  • C is the number of channels of the feature
  • Q p represents the query vector during prototype calibration
  • K p represents the key vector during prototype calibration
  • V p represents the value vector during prototype calibration.
  • the feature calibration module obtains the relationship matrix between input features through Qp and Kp , and combines Vp to obtain the attention feature to strengthen the connection between the prototype and the query set features; it extracts the information between feature channels through Qp , adjusts the attention feature, and obtains a better feature distribution.
  • step s4 comprises the following steps:
  • the calibration prototype feature and the calibration query feature are used as nodes of a graph
  • the prototype labels of the calibration prototype features are continuously propagated on the graph to obtain the corresponding label values of each node;
  • a Gaussian similarity matrix is used to represent the similarity between the nodes, and normalized to achieve label propagation
  • Extract node information corresponding to the query set feature normalize the node information to obtain a prediction corresponding to each query set feature, and realize semantic segmentation of the query set point cloud.
  • the calibration prototype features and the calibration query features are used as nodes v of the graph, and the prototype labels are continuously propagated on the graph to obtain the corresponding label values of each node.
  • the propagation process depends on the adjacency matrix of the nodes in the graph.
  • the node information corresponding to the query set features is extracted and normalized to obtain the prediction corresponding to each query set feature.
  • the semantic segmentation of the query set point cloud is achieved.
  • the loss function includes the loss between the pseudo label and the real label during label propagation and the loss between the final predicted value and the real label.
  • the present invention proposes a scheme for prototype expansion and feature calibration.
  • the prototype of the support set data has misjudgment results when assigning labels to the query set data.
  • Embodiment 2 is a diagrammatic representation of Embodiment 1:
  • the present application also provides a small sample point cloud semantic segmentation network using the above semantic segmentation method, including a sequentially connected feature extractor, a prototype amplification module, a feature calibration module and a label propagation module;
  • the feature extractor extracts features from the support set point cloud and the query set point cloud, and outputs corresponding support set features and query set features;
  • the prototype augmentation module combines the support set features and the query set features to obtain augmented multiple prototypes
  • the feature calibration module exchanges information between the multiple prototypes and the query set features to obtain calibration prototype features and calibration query features
  • the label propagation module measures the distance between the calibration query feature and the calibration prototype feature to perform label propagation, thereby achieving semantic segmentation of the query set point cloud.
  • the prototype amplification module includes a first feature aggregation unit, a label assignment unit, a second feature aggregation unit and a combination unit connected in sequence;
  • the first feature aggregation unit performs feature aggregation on the support set features based on the support set mask to obtain multiple prototypes of the support set;
  • the label assignment unit assigns each query set feature a label of the multi-prototype of the support set closest to it according to the Euclidean distance from the query set feature to the multi-prototype of the support set, thereby obtaining a pseudo label of the query set feature;
  • the second feature aggregation unit performs feature aggregation on the query set features based on the pseudo labels of the query set to obtain a pseudo prototype
  • the combining unit combines the multiple prototypes of the support set with the pseudo-prototype to obtain an augmented multiple prototype
  • the feature calibration module realizes the information flow between prototype information and query set features through two shared residual cross-attention modules.
  • the feature calibration module contains two shared residual cross attention modules to achieve information exchange between the prototype feature p and the query set feature fq .
  • the residual cross attention module uses the attention mechanism to extract the relationship matrix between the input features and uses the query vector to extract channel information. Taking prototype calibration as an example, its calculation method is as follows:
  • Q p k 1 (p)
  • K p k 2 (f q )
  • V p k 1 (p)
  • C is the number of channels of the feature.
  • Q p represents the query vector during prototype calibration
  • K p represents the key vector during prototype calibration
  • V p represents the value vector during prototype calibration.
  • the label propagation module uses the calibration prototype features and the calibration query features as the nodes v of the graph, and continuously propagates the prototype labels on the graph to obtain the corresponding label values of each node. Its propagation process depends on the adjacency matrix of the nodes in the graph. We use the Gaussian similarity matrix to represent the similarity between nodes and normalize it to achieve label propagation. Then, the node information corresponding to the query set features is extracted and normalized to obtain the predictions corresponding to each query and feature.
  • the loss function includes the loss of pseudo labels and true labels during label propagation and the loss of the final predicted value and true label.
  • the present application also provides a storage medium, which stores a program file that can implement the above-mentioned small sample point cloud semantic segmentation method.
  • Embodiment 4 is a diagrammatic representation of Embodiment 4:
  • the present application also provides a processor, which is used to run a program, wherein the program executes the above-mentioned small sample point cloud semantic segmentation method when running.
  • the small sample point cloud semantic segmentation method can be implemented by corresponding hardware or software units.
  • Each unit can be an independent software or hardware unit, or can be integrated into a software or hardware unit. This is not intended to limit the present application.
  • the specific implementation of each unit can refer to the description of the first embodiment.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Geometry (AREA)
  • Computer Graphics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Machine Translation (AREA)

Abstract

The present application is applicable to the technical field of point cloud semantic segmentation. Provided are a few-shot point cloud semantic segmentation method, and a network, a storage medium and a processor. In the few-shot point cloud semantic segmentation method, extracted pseudo prototype features adapted to query set data are propagated by using labels, such that prototype features adapted to the query set data are acquired, and feature calibration is performed by means of extracting the relationship between prototypes and the query set data. During prototype augmentation, distribution information of the query set data and prototype information of a support set are effectively used. The adaptability of the prototypes to the query set data is further improved. Therefore, the present invention can obtain prototype features adapted to a query set, thereby implementing effective segmentation of a point cloud scene, and reducing the impact of an erroneous judgment result on augmented prototypes.

Description

一种小样本点云语义分割方法、网络及存储介质和处理器Small sample point cloud semantic segmentation method, network, storage medium and processor 技术领域Technical Field
本申请属于点云语义分割技术领域，尤其涉及一种小样本点云语义分割方法、网络及存储介质和处理器。The present application belongs to the technical field of point cloud semantic segmentation, and in particular, relates to a small sample point cloud semantic segmentation method, network, storage medium and processor.
背景技术Background technique
目前的小样本点云语义分割技术主要利用带标注的支持集数据的原型特征对查询集数据进行标签传播以得到相应的点云标签，但因为标注数据的数量较少，支持集数据与查询集数据存在偏差，因此单纯利用支持集数据得到的决策边界难以在查询集数据上实现准确的分割。现有技术存在不足The current small sample point cloud semantic segmentation technology mainly uses the prototype features of the annotated support set data to propagate labels on the query set data to obtain the corresponding point cloud labels. However, due to the small amount of annotated data, there is a deviation between the support set data and the query set data. Therefore, it is difficult to achieve accurate segmentation on the query set data by simply using the decision boundary obtained from the support set data.
发明内容Summary of the invention
本申请的目的在于提供一种小样本点云语义分割方法、网络及存储介质和处理器，通过灵活配合次级用户所赋予的多重功能并调控相应的***资源，为认知***的信息安全场景推导可达速率区域，解决授权用户与认知用户建立安全共存关系的技术问题。The purpose of this application is to provide a small sample point cloud semantic segmentation method, network and storage medium and processor, which can flexibly cooperate with the multiple functions assigned by secondary users and regulate the corresponding system resources to derive the reachable rate area for the information security scenario of the cognitive system and solve the technical problem of establishing a secure coexistence relationship between authorized users and cognitive users.
一方面，本申请提供了一种小样本点云语义分割方法，包括下述步骤：On the one hand, the present application provides a small sample point cloud semantic segmentation method, comprising the following steps:
s1.将支持集点云和查询集点云输入特征提取器得到相应的支持集特征和查询集特征；s1. Input the support set point cloud and the query set point cloud into the feature extractor to obtain the corresponding support set features and query set features;
s2.将所述支持集特征和所述查询集特征输入原型扩增模块以获得扩增的多原型；s2. Inputting the support set features and the query set features into the prototype expansion module to obtain an expanded multi-prototype;
s3.将所述扩增的多原型和所述查询集特征输入到特征校准模块以获得校 准原型特征和校准查询特征；s3. inputting the amplified multiple prototypes and the query set features into a feature calibration module to obtain calibration prototype features and calibration query features;
s4.通过标签传播模块度量所述校准查询特征与所述校准原型特征之间的距离，进行标签传播，实现对查询集点云的语义分割。s4. The distance between the calibration query feature and the calibration prototype feature is measured by a label propagation module, and label propagation is performed to achieve semantic segmentation of the query set point cloud.
另一方面，本申请还提供了一种采用如上述语义分割方法的小样本点云语义分割网络，包括顺序连接的特征提取器、原型扩增模块、特征校准模块和标签传播模块；On the other hand, the present application also provides a small sample point cloud semantic segmentation network using the above semantic segmentation method, including a sequentially connected feature extractor, a prototype amplification module, a feature calibration module and a label propagation module;
所述特征提取器对支持集点云和查询集点云进行特征提取，输出相应的支持集特征和查询集特征；The feature extractor extracts features from the support set point cloud and the query set point cloud, and outputs corresponding support set features and query set features;
所述原型扩增模块将所述支持集特征和所述查询集特征进行组合获得扩增的多原型；The prototype augmentation module combines the support set features and the query set features to obtain augmented multiple prototypes;
所述特征校准模块对所述多原型和所述查询集特征进行信息交流，以获得校准原型特征和校准查询特征；The feature calibration module exchanges information between the multiple prototypes and the query set features to obtain calibration prototype features and calibration query features;
所述标签传播模块度量所述校准查询特征与所述校准原型特征之间的距离进行标签传播，实现对所述查询集点云的语义分割。The label propagation module measures the distance between the calibration query feature and the calibration prototype feature to perform label propagation, thereby achieving semantic segmentation of the query set point cloud.
另一方面，本申请还提供了一种存储介质，所述存储介质存储有能够实现上述的小样本点云语义分割方法的程序文件。On the other hand, the present application also provides a storage medium, which stores a program file that can implement the above-mentioned small sample point cloud semantic segmentation method.
另一方面，本申请还提供了一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行上述的小样本点云语义分割方法。On the other hand, the present application also provides a processor, which is used to run a program, wherein the program executes the above-mentioned small sample point cloud semantic segmentation method when running.
本申请提出了一种小样本点云语义分割方法、网络及存储介质和处理器。本申请利用标签传播提取适应于查询集数据的伪原型特征，从而获取适应于查询集数据的原型特征，并通过提取原型与查询集数据之间的关系进行特征校准。原型扩充有效利用了查询集数据的分布信息和支持集的原型信息。进一步提高原型对查询集数据的适应性。因此，本发明可以得到适应于查询集的原型特征， 实现对点云场景的有效分割。The present application proposes a small sample point cloud semantic segmentation method, network, storage medium and processor. The present application uses label propagation to extract pseudo-prototype features that are suitable for query set data, thereby obtaining prototype features that are suitable for query set data, and performs feature calibration by extracting the relationship between the prototype and the query set data. Prototype expansion effectively utilizes the distribution information of the query set data and the prototype information of the support set. The adaptability of the prototype to the query set data is further improved. Therefore, the present invention can obtain prototype features that are suitable for the query set and realize effective segmentation of the point cloud scene.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
为了更清楚地说明本申请实施例的技术方案，下面将对本申请实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面所描述的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
图1是本申请小样本点云语义分割方法的流程示意图；FIG1 is a schematic diagram of the process of the semantic segmentation method of a small sample point cloud of the present application;
图2是本申请原型扩增模块的架构示意图；FIG2 is a schematic diagram of the architecture of the prototype amplification module of the present application;
图3是本申请特征校准模块的架构示意图；FIG3 is a schematic diagram of the architecture of the feature calibration module of the present application;
图4本申请小样本点云语义分割方法的主要步骤示意图。FIG4 is a schematic diagram of the main steps of the small sample point cloud semantic segmentation method of this application.
具体实施方式Detailed ways
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图1-4及实施例，对本申请进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with Figures 1-4 and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
以下结合具体实施例对本申请的具体实现进行详细描述：The specific implementation of this application is described in detail below in conjunction with specific embodiments:
实施例一：Embodiment 1:
图1-4示出了本申请实施例一提供的小样本点云语义分割方法的实现流程，为了便于说明，仅示出了与本申请实施例相关的部分，详述如下：FIG1-4 shows the implementation process of the small sample point cloud semantic segmentation method provided in the first embodiment of the present application. For the sake of convenience, only the part related to the embodiment of the present application is shown, which is described in detail as follows:
一方面，本申请提供了一种小样本点云语义分割方法，包括下述步骤：On the one hand, the present application provides a small sample point cloud semantic segmentation method, comprising the following steps:
s1.将支持集点云和查询集点云输入特征提取器得到相应的支持集特征和查询集特征；s1. Input the support set point cloud and the query set point cloud into the feature extractor to obtain the corresponding support set features and query set features;
s2.将所述支持集特征和所述查询集特征输入原型扩增模块以获得扩增的多原型；s2. Inputting the support set features and the query set features into the prototype expansion module to obtain an expanded multi-prototype;
s3.将所述扩增的多原型和所述查询集特征输入到特征校准模块以获得校准原型特征和校准查询特征；s3. The amplified multi-prototype and the query set features are input into a feature calibration module to obtain calibration prototype features and calibration query features;
s4.通过标签传播模块度量所述校准查询特征与所述校准原型特征之间的距离，进行标签传播，实现对查询集点云的语义分割。s4. The distance between the calibration query feature and the calibration prototype feature is measured by a label propagation module, and label propagation is performed to achieve semantic segmentation of the query set point cloud.
具体的，本发明的方案流程如图1所示。支持集数据和查询集数据经过共享的特征提取器得到相应的支持集特征f s和查询集特征f q。之后，我们将支持集特征f s和查询集特征f q输入到原型扩增模块以获得多原型p。同时将扩增后的多原型p和查询集特征f q输入到特征校准模块以获得校准原型特征
Figure PCTCN2022135087-appb-000001
和校准查询特征
Figure PCTCN2022135087-appb-000002
最后通过标签传播实现对点云的语义分割。 Specifically, the solution flow of the present invention is shown in FIG1 . The support set data and the query set data are passed through a shared feature extractor to obtain the corresponding support set features f s and query set features f q . Afterwards, we input the support set features f s and query set features f q into the prototype expansion module to obtain multiple prototypes p. At the same time, the expanded multiple prototypes p and query set features f q are input into the feature calibration module to obtain the calibration prototype features
Figure PCTCN2022135087-appb-000001
and calibration query features
Figure PCTCN2022135087-appb-000002
Finally, semantic segmentation of point cloud is achieved through label propagation.
进一步的，所述步骤s2包括以下步骤：Furthermore, the step s2 comprises the following steps:
s21.所述支持集特征通过与支持集掩膜进行特征聚合获得支持集的多原型；s21. The support set features are aggregated with the support set mask to obtain multiple prototypes of the support set;
s22.依据查询集特征到所述支持集的多原型的欧氏距离，赋予每个所述查询集特征，距离其最近的所述支持集的多原型的标签；得到查询集特征的伪标签；s22. According to the Euclidean distance from the query set feature to the multi-prototype of the support set, assign each query set feature the label of the multi-prototype of the support set closest to it; obtain the pseudo label of the query set feature;
s23.所述查询集特征通过与所述查询集的伪标签进行特征聚合得到伪原型；s23. The query set features are aggregated with the pseudo labels of the query set to obtain a pseudo prototype;
s24.所述支持集的多原型与所述伪原型组合得到扩增的多原型。s24. The multiple prototypes of the support set are combined with the pseudo-prototype to obtain the expanded multiple prototypes.
原型扩增模块的结构如图2所示。我们将支持集特征和查询集特征输入到 原型扩增模块，支持集特征通过特征聚合获得支持集的多原型。通过特征聚合，我们得到了相应的原型特征。之后，依据查询集特征到支持集的多原型的欧氏距离，我们赋予每个查询集特征，距离其最近的支持集的多原型的标签，得到了查询集特征的伪标签。之后，查询集特征经过相同的特征聚合得到伪原型。通过组合支持集的多原型与伪原型，我们得到扩增的多原型。The structure of the prototype augmentation module is shown in Figure 2. We input the support set features and query set features into the prototype augmentation module, and the support set features are aggregated to obtain the multi-prototypes of the support set. Through feature aggregation, we obtain the corresponding prototype features. After that, according to the Euclidean distance from the query set features to the multi-prototypes of the support set, we assign each query set feature the label of the multi-prototype of the support set closest to it, and obtain the pseudo-label of the query set feature. After that, the query set features are aggregated in the same way to obtain the pseudo-prototype. By combining the multi-prototypes of the support set with the pseudo-prototype, we obtain the augmented multi-prototype.
进一步的，所述特征聚合包括以下步骤：Furthermore, the feature aggregation includes the following steps:
j1.依据标签将同类特征提取出来；j1. Extract similar features based on labels;
j2.对所述同类特征执行最远点采样得到初始聚类中心；j2. Perform the farthest point sampling on the same features to obtain the initial cluster center;
j3.根据特征到所述初始聚类中心的距离将其分配到最近的聚类中心；j3. Assigning the feature to the nearest cluster center according to its distance to the initial cluster center;
j4.对属于每个所述聚类中心的特征取平均作为最终的原型特征。j4. Take the average of the features belonging to each cluster center as the final prototype feature.
进一步的，所述特征校准模块通过两个共享的残差交叉注意力模块实现所述多原型的信息与所述查询集特征之间的信息流通，进而输出所述校准原型特征和所述校准查询特征。Furthermore, the feature calibration module realizes the information flow between the multi-prototype information and the query set features through two shared residual cross-attention modules, and then outputs the calibrated prototype features and the calibrated query features.
进一步的，所述残差交叉注意力模块通过三个卷积核得到注意力特征和查询向量特征，并采用注意力机制提取输入特征之间的关系矩阵，利用查询向量实现通道信息的提取。Furthermore, the residual cross attention module obtains attention features and query vector features through three convolution kernels, and uses the attention mechanism to extract the relationship matrix between input features, and uses the query vector to extract channel information.
具体的，如附图3所示，图3(a)部分为特征校准模块的具体架构，通过两个共享的残差注意力模块实现原型信息与查询集特征之间的信息流通；图3(b)部分为残差交叉注意力模块的具体机构，通过三个卷积核得到注意力特征和查询向量特征。Specifically, as shown in Figure 3, Figure 3(a) is the specific architecture of the feature calibration module, which realizes the information flow between the prototype information and the query set features through two shared residual attention modules; Figure 3(b) is the specific structure of the residual cross attention module, which obtains the attention features and query vector features through three convolution kernels.
我们将扩增后的原型输入到特征校准模块，如附图3所示。特征校准模块包含两个共享的残差交叉注意力模块，以此实现原型特征p与查询集特征f q之间的信息交流。残差交叉注意力模块采用注意力机制提取输入特征之间的关系 矩阵，并利用查询向量实现通道信息的提取。以原型校准为例，其计算方法如下： We input the amplified prototype into the feature calibration module, as shown in Figure 3. The feature calibration module contains two shared residual cross attention modules to achieve information exchange between the prototype feature p and the query set feature fq . The residual cross attention module uses the attention mechanism to extract the relationship matrix between the input features and uses the query vector to extract channel information. Taking prototype calibration as an example, the calculation method is as follows:
Figure PCTCN2022135087-appb-000003
Figure PCTCN2022135087-appb-000003
其中Q p＝k 1(p)，K p＝k 2(f q)，V p＝k 1(p)，C是特征的通道数。Q p表示原型校准时的查询向量，K p表示原型校准时的键向量，V p表示原型校准时的值向量。 Where Q p = k 1 (p), K p = k 2 (f q ), V p = k 1 (p), C is the number of channels of the feature, Q p represents the query vector during prototype calibration, K p represents the key vector during prototype calibration, and V p represents the value vector during prototype calibration.
特征校准模块通过Q p，K p得到输入特征之间的关系矩阵，并结合V p得到注意力特征，加强原型与查询集特征之间的联系；通过Q p提取特征通道间的信息，对注意力特征进行调整，获得更好的特征分布。 The feature calibration module obtains the relationship matrix between input features through Qp and Kp , and combines Vp to obtain the attention feature to strengthen the connection between the prototype and the query set features; it extracts the information between feature channels through Qp , adjusts the attention feature, and obtains a better feature distribution.
经过特征校准模块后，我们得到校准后的原型特征
Figure PCTCN2022135087-appb-000004
与查询集特征
Figure PCTCN2022135087-appb-000005
After the feature calibration module, we get the calibrated prototype features
Figure PCTCN2022135087-appb-000004
With queryset features
Figure PCTCN2022135087-appb-000005
进一步的，所述步骤s4包括以下步骤：Further, the step s4 comprises the following steps:
s41.将所述校准原型特征与所述校准查询特征作为图的结点；s41. The calibration prototype feature and the calibration query feature are used as nodes of a graph;
s42.通过所述校准原型特征的原型标签在图上不断传播，得到各节点相应的标签值；s42. The prototype labels of the calibration prototype features are continuously propagated on the graph to obtain the corresponding label values of each node;
s43.采用高斯相似度矩阵来表示各个所述结点之间的相似度，并将其归一化以实现标签传播；s43. A Gaussian similarity matrix is used to represent the similarity between the nodes, and normalized to achieve label propagation;
s44.提取与所述查询集特征对应的节点信息，归一化所述节点信息得到每个所述查询集特征对应的预测，实现对查询集点云的语义分割。s44. Extract node information corresponding to the query set feature, normalize the node information to obtain a prediction corresponding to each query set feature, and realize semantic segmentation of the query set point cloud.
具体的，将校准原型特征与校准查询特征作为图的结点v，通过原型标签在图上不断传播得到各节点相应的标签值。其传播过程依赖于图中结点的邻接矩阵。我们采用高斯相似度矩阵来表示结点之间的相似度，并将其归一化以实现标签传播。之后提取与查询集特征对应的节点信息，并归一化得到每个查询集特征对应的预测。实现对查询集点云的语义分割。Specifically, the calibration prototype features and the calibration query features are used as nodes v of the graph, and the prototype labels are continuously propagated on the graph to obtain the corresponding label values of each node. The propagation process depends on the adjacency matrix of the nodes in the graph. We use the Gaussian similarity matrix to represent the similarity between nodes and normalize it to achieve label propagation. Then, the node information corresponding to the query set features is extracted and normalized to obtain the prediction corresponding to each query set feature. The semantic segmentation of the query set point cloud is achieved.
这其中，损失函数包括在标签传播时伪标签与真实标签的损失及最后预测值与真实标签的损失。Among them, the loss function includes the loss between the pseudo label and the real label during label propagation and the loss between the final predicted value and the real label.
本发明提出了一种原型扩增和特征校准的方案。针对少量支持集数据的原型在查询集数据上难以获得优秀的决策边界的问题，我们提出利用查询集数据对原型进行扩充。同时考虑到扩充原型依赖于支持集数据的原型，而支持集数据的原型在对查询集数据赋予标签时存在误判结果。为降低误判结果对扩充原型的影响，我们提出利用原型与查询集数据之间关系来对特征进行校准。为进一步降低误判数据的影响，我们在原有分割损失函数的基础上增加了标签赋予的损失函数。The present invention proposes a scheme for prototype expansion and feature calibration. In view of the problem that it is difficult for a prototype with a small amount of support set data to obtain an excellent decision boundary on the query set data, we propose to expand the prototype using the query set data. At the same time, considering that the expanded prototype depends on the prototype of the support set data, and the prototype of the support set data has misjudgment results when assigning labels to the query set data. In order to reduce the impact of misjudgment results on the expanded prototype, we propose to calibrate the features using the relationship between the prototype and the query set data. In order to further reduce the impact of misjudged data, we added a label assignment loss function on the basis of the original segmentation loss function.
实施例二：Embodiment 2:
另一方面，本申请还提供了一种采用如上述语义分割方法的小样本点云语义分割网络，包括顺序连接的特征提取器、原型扩增模块、特征校准模块和标签传播模块；On the other hand, the present application also provides a small sample point cloud semantic segmentation network using the above semantic segmentation method, including a sequentially connected feature extractor, a prototype amplification module, a feature calibration module and a label propagation module;
所述特征提取器对支持集点云和查询集点云进行特征提取，输出相应的支持集特征和查询集特征；The feature extractor extracts features from the support set point cloud and the query set point cloud, and outputs corresponding support set features and query set features;
所述原型扩增模块将所述支持集特征和所述查询集特征进行组合获得扩增的多原型；The prototype augmentation module combines the support set features and the query set features to obtain augmented multiple prototypes;
所述特征校准模块对所述多原型和所述查询集特征进行信息交流，以获得校准原型特征和校准查询特征；The feature calibration module exchanges information between the multiple prototypes and the query set features to obtain calibration prototype features and calibration query features;
所述标签传播模块度量所述校准查询特征与所述校准原型特征之间的距离进行标签传播，实现对所述查询集点云的语义分割。The label propagation module measures the distance between the calibration query feature and the calibration prototype feature to perform label propagation, thereby achieving semantic segmentation of the query set point cloud.
进一步的，所述原型扩增模块包括顺序连接的第一特征聚合单元、标签赋予单元、第二特征聚合单元和组合单元；Further, the prototype amplification module includes a first feature aggregation unit, a label assignment unit, a second feature aggregation unit and a combination unit connected in sequence;
所述第一特征聚合单元基于支持集掩膜对所述支持集特征进行特征聚合，获得支持集的多原型；The first feature aggregation unit performs feature aggregation on the support set features based on the support set mask to obtain multiple prototypes of the support set;
所述标签赋予单元依据查询集特征到所述支持集的多原型的欧氏距离，赋予每个所述查询集特征，距离其最近的所述支持集的多原型的标签；得到查询集特征的伪标签；The label assignment unit assigns each query set feature a label of the multi-prototype of the support set closest to it according to the Euclidean distance from the query set feature to the multi-prototype of the support set, thereby obtaining a pseudo label of the query set feature;
所述第二特征聚合单元基于所述查询集的伪标签对所述查询集特征进行特征聚合，得到伪原型；The second feature aggregation unit performs feature aggregation on the query set features based on the pseudo labels of the query set to obtain a pseudo prototype;
所述组合单元将所述支持集的多原型与所述伪原型组合得到扩增的多原型；The combining unit combines the multiple prototypes of the support set with the pseudo-prototype to obtain an augmented multiple prototype;
所述特征校准模块通过两个共享的残差交叉注意力模块实现原型信息与查询集特征之间的信息流通。The feature calibration module realizes the information flow between prototype information and query set features through two shared residual cross-attention modules.
具体的，特征校准模块包含两个共享的残差交叉注意力模块，以此实现原型特征p与查询集特征f q之间的信息交流。残差交叉注意力模块采用注意力机制提取输入特征之间的关系矩阵，并利用查询向量实现通道信息的提取。以原型校准为例，其计算方法如下： Specifically, the feature calibration module contains two shared residual cross attention modules to achieve information exchange between the prototype feature p and the query set feature fq . The residual cross attention module uses the attention mechanism to extract the relationship matrix between the input features and uses the query vector to extract channel information. Taking prototype calibration as an example, its calculation method is as follows:
Figure PCTCN2022135087-appb-000006
Figure PCTCN2022135087-appb-000006
其中Q p＝k 1(p)，K p＝k 2(f q)，V p＝k 1(p)，C是特征的通道数。Q p表示原型校准时的查询向量，K p表示原型校准时的键向量，V p表示原型校准时的值向量。。 Where Q p = k 1 (p), K p = k 2 (f q ), V p = k 1 (p), C is the number of channels of the feature. Q p represents the query vector during prototype calibration, K p represents the key vector during prototype calibration, and V p represents the value vector during prototype calibration.
经过特征校准模块后，我们得到校准后的原型特征
Figure PCTCN2022135087-appb-000007
与查询集特征
Figure PCTCN2022135087-appb-000008
After the feature calibration module, we get the calibrated prototype features
Figure PCTCN2022135087-appb-000007
With queryset features
Figure PCTCN2022135087-appb-000008
标签传播模块则将校准原型特征与校准查询特征作为图的结点v，通过原型标签在图上不断传播得到各节点相应的标签值。其传播过程依赖于图中结点的邻接矩阵。我们采用高斯相似度矩阵来表示结点之间的相似度，并将其归一 化以实现标签传播。之后提取与查询集特征对应的节点信息，并归一化得到每个查询及特征对应的预测。损失函数包括在标签传播时伪标签与真实标签的损失及最后预测值与真实标签的损失。The label propagation module uses the calibration prototype features and the calibration query features as the nodes v of the graph, and continuously propagates the prototype labels on the graph to obtain the corresponding label values of each node. Its propagation process depends on the adjacency matrix of the nodes in the graph. We use the Gaussian similarity matrix to represent the similarity between nodes and normalize it to achieve label propagation. Then, the node information corresponding to the query set features is extracted and normalized to obtain the predictions corresponding to each query and feature. The loss function includes the loss of pseudo labels and true labels during label propagation and the loss of the final predicted value and true label.
实施例三：Embodiment three:
另一方面，本申请还提供了一种存储介质，所述存储介质存储有能够实现上述的小样本点云语义分割方法的程序文件。On the other hand, the present application also provides a storage medium, which stores a program file that can implement the above-mentioned small sample point cloud semantic segmentation method.
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，所述的程序可以存储于一计算机可读取存储介质中，所述的存储介质，如ROM/RAM、磁盘、光盘等。A person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, such as ROM/RAM, disk, CD-ROM, etc.
实施例四：Embodiment 4:
另一方面，本申请还提供了一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行上述的小样本点云语义分割方法。On the other hand, the present application also provides a processor, which is used to run a program, wherein the program executes the above-mentioned small sample point cloud semantic segmentation method when running.
在本申请实施例中，该小样本点云语义分割方法可由相应的硬件或软件单元实现，各单元可以为独立的软、硬件单元，也可以集成为一个软、硬件单元，在此不用以限制本申请。各单元的具体实施方式可参考实施例一的描述，在此In the embodiment of the present application, the small sample point cloud semantic segmentation method can be implemented by corresponding hardware or software units. Each unit can be an independent software or hardware unit, or can be integrated into a software or hardware unit. This is not intended to limit the present application. The specific implementation of each unit can refer to the description of the first embodiment.
以上所述仅为本申请的较佳实施例而已，并不用以限制本申请，凡在本申请的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本申请的保护范围之内不再赘述。The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application and will not be repeated.

Claims (10)

  1. 一种小样本点云语义分割方法，其特征在于，包括下述步骤：A small sample point cloud semantic segmentation method, characterized by comprising the following steps:
    s1.将支持集点云和查询集点云输入特征提取器得到相应的支持集特征和查询集特征；s1. Input the support set point cloud and the query set point cloud into the feature extractor to obtain the corresponding support set features and query set features;
    s2.将所述支持集特征和所述查询集特征输入原型扩增模块以获得扩增的多原型；s2. Inputting the support set features and the query set features into the prototype expansion module to obtain an expanded multi-prototype;
    s3.将所述扩增的多原型和所述查询集特征输入到特征校准模块以获得校准原型特征和校准查询特征；s3. The amplified multi-prototype and the query set features are input into a feature calibration module to obtain calibration prototype features and calibration query features;
    s4.通过标签传播模块度量所述校准查询特征与所述校准原型特征之间的距离，进行标签传播，实现对查询集点云的语义分割。s4. The distance between the calibration query feature and the calibration prototype feature is measured by a label propagation module, and label propagation is performed to achieve semantic segmentation of the query set point cloud.
  2. 如权利要求1所述的语义分割方法，其特征在于，所述步骤s2包括以下步骤：The semantic segmentation method according to claim 1, characterized in that step s2 comprises the following steps:
    s21.所述支持集特征通过与支持集掩膜进行特征聚合获得支持集的多原型；s21. The support set features are aggregated with the support set mask to obtain multiple prototypes of the support set;
    s22.依据查询集特征到所述支持集的多原型的欧氏距离，赋予每个所述查询集特征，距离其最近的所述支持集的多原型的标签；得到查询集特征的伪标签；s22. According to the Euclidean distance from the query set feature to the multi-prototype of the support set, assign each query set feature the label of the multi-prototype of the support set closest to it; obtain the pseudo label of the query set feature;
    s23.所述查询集特征通过与所述查询集的伪标签进行特征聚合得到伪原型；s23. The query set features are aggregated with the pseudo labels of the query set to obtain a pseudo prototype;
    s24.所述支持集的多原型与所述伪原型组合得到扩增的多原型。s24. The multiple prototypes of the support set are combined with the pseudo-prototype to obtain the expanded multiple prototypes.
  3. 如权利要求2所述的语义分割方法，其特征在于，所述特征聚合包括以下步骤：The semantic segmentation method according to claim 2, wherein the feature aggregation comprises the following steps:
    j1.依据标签将同类特征提取出来；j1. Extract similar features based on labels;
    j2.对所述同类特征执行最远点采样得到初始聚类中心；j2. Perform the farthest point sampling on the same features to obtain the initial cluster center;
    j3.根据特征到所述初始聚类中心的距离将其分配到最近的聚类中心；j3. Assigning the feature to the nearest cluster center according to its distance to the initial cluster center;
    j4.对属于每个所述聚类中心的特征取平均作为最终的原型特征。j4. Take the average of the features belonging to each cluster center as the final prototype feature.
  4. 如权利要求1所述的语义分割方法，其特征在于，所述特征校准模块通过两个共享的残差交叉注意力模块实现所述多原型的信息与所述查询集特征之间的信息流通，进而输出所述校准原型特征和所述校准查询特征。The semantic segmentation method as described in claim 1 is characterized in that the feature calibration module realizes the information flow between the multi-prototype information and the query set features through two shared residual cross-attention modules, and then outputs the calibrated prototype features and the calibrated query features.
  5. 如权利要求4所述的语义分割方法，其特征在于，所述残差交叉注意力模块通过三个卷积核得到注意力特征和查询向量特征，并采用注意力机制提取输入特征之间的关系矩阵，利用查询向量实现通道信息的提取。The semantic segmentation method as described in claim 4 is characterized in that the residual cross-attention module obtains attention features and query vector features through three convolution kernels, and uses the attention mechanism to extract the relationship matrix between input features, and uses the query vector to realize the extraction of channel information.
  6. 如权利要求5所述的语义分割方法，其特征在于，所述步骤s4包括以下步骤：The semantic segmentation method according to claim 5, characterized in that step s4 comprises the following steps:
    s41.将所述校准原型特征与所述校准查询特征作为图的结点；s41. The calibration prototype feature and the calibration query feature are used as nodes of a graph;
    s42.通过所述校准原型特征的原型标签在图上不断传播，得到各节点相应的标签值；s42. The prototype labels of the calibration prototype features are continuously propagated on the graph to obtain the corresponding label values of each node;
    s43.采用高斯相似度矩阵来表示各个所述结点之间的相似度，并将其归一化以实现标签传播；s43. A Gaussian similarity matrix is used to represent the similarity between the nodes, and normalized to achieve label propagation;
    s44.提取与所述查询集特征对应的节点信息，归一化所述节点信息得到每个所述查询集特征对应的预测，实现对查询集点云的语义分割。s44. Extract node information corresponding to the query set feature, normalize the node information to obtain a prediction corresponding to each query set feature, and realize semantic segmentation of the query set point cloud.
  7. 一种采用如权利要求1至6任意一项所述语义分割方法的小样本点云语义分割网络，其特征在于，包括顺序连接的特征提取器、原型扩增模块、特征校准模块和标签传播模块；A small sample point cloud semantic segmentation network using the semantic segmentation method according to any one of claims 1 to 6, characterized in that it comprises a feature extractor, a prototype amplification module, a feature calibration module and a label propagation module connected in sequence;
    所述特征提取器对支持集点云和查询集点云进行特征提取，输出相应的支持集特征和查询集特征；The feature extractor extracts features from the support set point cloud and the query set point cloud, and outputs corresponding support set features and query set features;
    所述原型扩增模块将所述支持集特征和所述查询集特征进行组合获得扩增的多原型；The prototype augmentation module combines the support set features and the query set features to obtain augmented multiple prototypes;
    所述特征校准模块对所述多原型和所述查询集特征进行信息交流，以获得校准原型特征和校准查询特征；The feature calibration module exchanges information between the multiple prototypes and the query set features to obtain calibration prototype features and calibration query features;
    所述标签传播模块度量所述校准查询特征与所述校准原型特征之间的距离进行标签传播，实现对所述查询集点云的语义分割。The label propagation module measures the distance between the calibration query feature and the calibration prototype feature to perform label propagation, thereby achieving semantic segmentation of the query set point cloud.
  8. 如权利要求6所述的语义分割方法，其特征在于，所述原型扩增模块包括顺序连接的第一特征聚合单元、标签赋予单元、第二特征聚合单元和组合单元；The semantic segmentation method according to claim 6, characterized in that the prototype expansion module comprises a first feature aggregation unit, a label assignment unit, a second feature aggregation unit and a combination unit connected in sequence;
    所述第一特征聚合单元基于支持集掩膜对所述支持集特征进行特征聚合，获得支持集的多原型；The first feature aggregation unit performs feature aggregation on the support set features based on the support set mask to obtain multiple prototypes of the support set;
    所述标签赋予单元依据查询集特征到所述支持集的多原型的欧氏距离，赋予每个所述查询集特征，距离其最近的所述支持集的多原型的标签；得到查询集特征的伪标签；The label assignment unit assigns each query set feature a label of the multi-prototype of the support set closest to it according to the Euclidean distance from the query set feature to the multi-prototype of the support set, thereby obtaining a pseudo label of the query set feature;
    所述第二特征聚合单元基于所述查询集的伪标签对所述查询集特征进行特征聚合，得到伪原型；The second feature aggregation unit performs feature aggregation on the query set features based on the pseudo labels of the query set to obtain a pseudo prototype;
    所述组合单元将所述支持集的多原型与所述伪原型组合得到扩增的多原型；The combining unit combines the multiple prototypes of the support set with the pseudo-prototype to obtain an augmented multiple prototype;
    所述特征校准模块通过两个共享的残差交叉注意力模块实现原型信息与查询集特征之间的信息流通。The feature calibration module realizes the information flow between prototype information and query set features through two shared residual cross-attention modules.
  9. 一种存储介质，其特征在于，所述存储介质存储有能够实现权利要求1至6中任意一项所述的小样本点云语义分割方法的程序文件。A storage medium, characterized in that the storage medium stores a program file capable of implementing the small sample point cloud semantic segmentation method described in any one of claims 1 to 6.
  10. 一种处理器，其特征在于，所述处理器用于运行程序，其中，所述程序运行时执行权利要求1至6中任意一项所述的小样本点云语义分割方法。A processor, characterized in that the processor is used to run a program, wherein the program, when running, executes the small sample point cloud semantic segmentation method described in any one of claims 1 to 6.
PCT/CN2022/135087 2022-11-29 2022-11-29 Few-shot point cloud semantic segmentation method, and network, storage medium and processor WO2024113169A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/135087 WO2024113169A1 (en) 2022-11-29 2022-11-29 Few-shot point cloud semantic segmentation method, and network, storage medium and processor

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/135087 WO2024113169A1 (en) 2022-11-29 2022-11-29 Few-shot point cloud semantic segmentation method, and network, storage medium and processor

Publications (1)

Publication Number Publication Date
WO2024113169A1 true WO2024113169A1 (en) 2024-06-06

Family

ID=91322780

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/135087 WO2024113169A1 (en) 2022-11-29 2022-11-29 Few-shot point cloud semantic segmentation method, and network, storage medium and processor

Country Status (1)

Country Link
WO (1) WO2024113169A1 (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10672129B1 (en) * 2019-03-22 2020-06-02 Lunit Inc. Method for semantic segmentation and apparatus thereof
CN113129311A (en) * 2021-03-10 2021-07-16 西北大学 Label optimization point cloud example segmentation method
CN113706487A (en) * 2021-08-17 2021-11-26 西安电子科技大学 Multi-organ segmentation method based on self-supervision characteristic small sample learning
CN114529728A (en) * 2022-04-25 2022-05-24 武汉工程大学 Small sample semantic segmentation method and device

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10672129B1 (en) * 2019-03-22 2020-06-02 Lunit Inc. Method for semantic segmentation and apparatus thereof
CN113129311A (en) * 2021-03-10 2021-07-16 西北大学 Label optimization point cloud example segmentation method
CN113706487A (en) * 2021-08-17 2021-11-26 西安电子科技大学 Multi-organ segmentation method based on self-supervision characteristic small sample learning
CN114529728A (en) * 2022-04-25 2022-05-24 武汉工程大学 Small sample semantic segmentation method and device

Similar Documents

Publication Publication Date Title
Yun et al. Khaos: An adversarial neural network DGA with high anti-detection ability
CN110084365A (en) A kind of service provider system and method based on deep learning
CN107784327A (en) A kind of personalized community discovery method based on GN
CN112860932B (en) Image retrieval method, device, equipment and storage medium for resisting malicious sample attack
CN111353554B (en) Method and device for predicting missing user service attributes
Zheng et al. GCN‐ETA: High‐Efficiency Encrypted Malicious Traffic Detection
Tong et al. BFSN: a novel method of encrypted traffic classification based on bidirectional flow sequence network
WO2019184325A1 (en) Community division quality evaluation method and system based on average mutual information
WO2024113169A1 (en) Few-shot point cloud semantic segmentation method, and network, storage medium and processor
CN113709152B (en) Antagonistic domain name generation model with high-resistance detection capability
Liu et al. WealthAdapt: A general network adaptation framework for small data tasks
Pradhan et al. Stationary distribution of an infinite-buffer batch-arrival and batch-service queue with random serving capacity and batch-size-dependent service
Pradhan On the distribution of an infinite-buffer queueing system with versatile bulk-service rule under batch-size-dependent service policy: M/G n (a, Y)/1
Deng et al. Multi‐step implicit Adams predictor‐corrector network for fire detection
Verma et al. A novel model to enhance the data security in cloud environment
Advaith et al. A proximity based community detection in temporal graphs
CN117014182A (en) Malicious traffic detection method and device based on LSTM
Kim et al. A graph embedding-based identity inference attack on blockchain systems
CN104463864B (en) Multistage parallel key frame cloud extracting method and system
Ye et al. Optimal locally private estimation under p loss for 1≤p≤2
CN116226852A (en) Mobile platform malicious software detection method and device based on multi-mode information fusion
Li et al. Mangling Rules Generation With Density-Based Clustering for Password Guessing
Ennadir et al. A Simple and Yet Fairly Effective Defense for Graph Neural Networks
CN110188692B (en) Enhanced cyclic cascading method for effective target rapid identification
Blanton et al. Secure and oblivious maximum bipartite matching size algorithm with applications to secure fingerprint identification

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22966767

Country of ref document: EP

Kind code of ref document: A1