CN113705689A - Training data acquisition method and abnormal behavior recognition network training method - Google Patents
Training data acquisition method and abnormal behavior recognition network training method Download PDFInfo
- Publication number
- CN113705689A CN113705689A CN202111006832.0A CN202111006832A CN113705689A CN 113705689 A CN113705689 A CN 113705689A CN 202111006832 A CN202111006832 A CN 202111006832A CN 113705689 A CN113705689 A CN 113705689A
- Authority
- CN
- China
- Prior art keywords
- data
- network
- training
- training data
- abnormal behavior
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/951—Indexing; Web crawling techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The specification provides a training data acquisition method and an abnormal behavior recognition network training method, aiming at a specific abnormal behavior, acquiring acquisition data containing the specific abnormal behavior, acquiring network data, acquiring action characteristics of the acquisition data and the network data, determining similar network data of a plurality of acquisition data according to the similarity of the action characteristics of each network data and the action characteristics of the acquisition data, and taking the similar network data and the acquisition data as positive sample training data of the specific abnormal behavior. Through the similarity of the action characteristics, training data which can be used as a positive sample is determined from a plurality of cheap network data, the training data acquisition efficiency is improved, and the acquisition of an abnormal behavior recognition network is further accelerated.
Description
Technical Field
The present disclosure relates to the field of computer application technologies, and in particular, to the field of machine learning technologies, and in particular, to a training data acquisition method and an abnormal behavior recognition network training method.
Background
In the scene of smart city management, abnormal behaviors (such as fighting a shelf, climbing and the like) are required to be automatically identified through video data, and harm to city safety harmony caused by the abnormal behaviors is prevented.
Automatic identification of abnormal behavior often requires training of the abnormal behavior identification network through labeled abnormal behavior data. And the collection and labeling of data often take a long time, so that the abnormal behavior recognition network cannot be trained quickly in a short time.
Disclosure of Invention
The specification provides a training data acquisition method and an abnormal behavior recognition network training method.
According to a first aspect of embodiments herein, there is provided a training data acquisition method, including:
acquiring network data and acquisition data containing specific abnormal behaviors;
acquiring action characteristics of each network data and each acquired data;
and selecting similar network data of the acquired data from the network data according to the similarity between the action characteristic of each network data and the action characteristic of the acquired data, and taking the acquired data and the similar network data as positive sample training data aiming at specific abnormal behaviors.
In some embodiments, the network data comprises: an internet published data set, and/or web crawler data, and/or data generated based on a virtual game engine.
Such network data is more comprehensive, so that training data is more diverse.
In some embodiments, the obtaining of the action characteristic of each of the network data and each of the collected data includes:
acquiring a backbone network;
extracting the action characteristic of each acquired data through the backbone network;
and acquiring the action characteristics of each network data which is stored in advance and extracted through the backbone network.
Through the backbone network, the action characteristics of the collected data can be acquired, the action characteristics of the network data are extracted in advance, and when the requirements of a plurality of abnormal behavior training are met, the action characteristics of the network data do not need to be acquired for many times, so that the training efficiency is improved.
In some embodiments, the selecting similar network data of the collected data according to the similarity between the action feature of each network data and the action feature of the collected data includes:
synthesizing the action characteristics of all acquired data into the characteristics of an acquired data center;
and selecting similar network data of the acquired data according to the similarity of the action characteristic of each network data and the characteristic of the acquired data center.
Therefore, the action characteristics of all the collected data are synthesized into the central characteristic, the central characteristic can reflect the characteristics of the abnormal behavior better, and the selected similar network data are more accurate.
In some embodiments, the selecting similar network data of the collected data according to the similarity between the action feature of each network data and the action feature of the collected data includes:
determining the quantity N of the similar network data to be acquired according to a preset quantity proportion and the quantity of the acquired data;
selecting N similar network data from the network data; and the similarity between the action characteristic of any one of the similar network data and the action characteristic of the collected data is not less than the similarity between the action characteristic of any one of the network data which is not selected and the action characteristic of the collected data.
Therefore, the selected proportion of the similar network data to the collected data meets a certain condition, so that the abnormal behavior recognition network is not biased by the similar network data, and a better effect is achieved on the collected data.
According to a second aspect of embodiments herein, there is provided an abnormal behavior recognition network training method, the method comprising:
acquiring training data, wherein the training data comprises positive sample training data and negative sample training data, and the positive sample training data is acquired based on the training data acquisition method;
and iteratively training the abnormal behavior recognition network through the positive sample training data and the negative sample training data until the loss of the network output is less than a preset first loss threshold value or the iteration times are greater than a preset time threshold value.
In some embodiments, the positive example training data comprises a first label and the negative example training data comprises a second label;
each iteration of training the abnormal behavior recognition network comprises:
obtaining a judgment result of each training data; the determination result is used for characterizing whether the training data comprises the specific abnormal behavior;
for each similar network data, if the judgment result of the similar network data is not in accordance with the label, discarding the similar network data;
obtaining the loss of the network output according to the judgment result of the training data which is not discarded and the label value;
and updating the weight of the abnormal behavior recognition network according to the loss of the network output.
Therefore, similar network data which are not similar to the collected data are discarded, and the deviation of the abnormal behavior recognition network carried by the data is avoided.
In some embodiments, in each iteration, before the obtaining the determination result of each of the training data, the method further comprises:
acquiring action characteristics of each training data according to a backbone network; the action feature is used to determine whether the training data includes the particular abnormal behavior;
in each iteration, after obtaining the loss of the network output according to the decision result and the label of the training data which are not discarded, the method further comprises:
and updating the weight of the backbone network according to the loss.
Therefore, by updating the backbone network, the characteristics of the backbone network that abnormal behaviors can be extracted are better, and the accuracy of the abnormal behavior identification network is improved.
In some embodiments, in each iteration, before the determining result for each similar network data does not match the label, the method further includes:
inputting each training data into a discriminator; the discriminator is used for judging whether the training data is the collected data;
for each similar network data, if the determination result of the network data does not match the label, discarding the network data, including:
and for each piece of training data, responding to the condition that a discriminator outputs the training data which is not the collected data and the judgment result of the training data is not consistent with the label of the training data, and discarding the training data.
Therefore, the input training data does not need to contain the characteristics of identifying whether the data is collected or not, and the complexity of the model is reduced.
According to a third aspect of embodiments herein, there is provided a training data acquisition apparatus including:
the data acquisition module is used for acquiring network data and acquisition data containing specific abnormal behaviors;
the action characteristic acquisition module is used for acquiring action characteristics of each piece of network data and each piece of acquired data;
and the training data selection module is used for selecting similar network data of the acquired data from the network data according to the similarity between the action characteristic of each network data and the action characteristic of the acquired data, and taking the acquired data and the similar network data as positive sample training data aiming at a specific abnormal behavior.
In some embodiments, the network data comprises: an internet published data set, and/or web crawler data, and/or data generated based on a virtual game engine.
Such network data is more comprehensive, so that training data is more diverse.
In some embodiments, the data acquisition module is specifically configured to: acquiring a backbone network; extracting the action characteristic of each acquired data through the backbone network; and acquiring the action characteristics of each network data which is stored in advance and extracted through the backbone network.
Through the backbone network, the action characteristics of the collected data can be acquired, the action characteristics of the network data are extracted in advance, and when the requirements of a plurality of abnormal behavior training are met, the action characteristics of the network data do not need to be acquired for many times, so that the training efficiency is improved.
In some embodiments, the training data selecting module selects similar network data of the collected data according to a similarity between an action feature of each of the network data and an action feature of the collected data, and includes: synthesizing the action characteristics of all acquired data into the characteristics of an acquired data center; and selecting similar network data of the acquired data according to the similarity of the action characteristic of each network data and the characteristic of the acquired data center.
Therefore, the action characteristics of all the collected data are synthesized into the central characteristic, the central characteristic can reflect the characteristics of the abnormal behavior better, and the selected similar network data are more accurate.
In some embodiments, the training data selecting module selects similar network data of the collected data according to a similarity between an action feature of each of the network data and an action feature of the collected data, and includes: determining the quantity N of the similar network data to be acquired according to a preset quantity proportion and the quantity of the acquired data; selecting N similar network data from the network data; and the similarity between the action characteristic of any one of the similar network data and the action characteristic of the collected data is not less than the similarity between the action characteristic of any one of the network data which is not selected and the action characteristic of the collected data.
Therefore, the selected proportion of the similar network data to the collected data meets a certain condition, so that the abnormal behavior recognition network is not biased by the similar network data, and a better effect is achieved on the collected data.
According to a fourth aspect of embodiments herein, there is provided an abnormal behavior recognition network training apparatus, the apparatus including:
the training data acquisition module is used for acquiring training data, wherein the training data comprises positive sample training data and negative sample training data, and the positive sample training data is acquired based on the training sample acquisition method;
and the network training module is used for iteratively training the abnormal behavior recognition network through the positive sample training data and the negative sample training data until the loss of the network output is less than a preset first loss threshold value or the iteration times is greater than a preset time threshold value.
In some embodiments, the positive example training data includes a first label and the negative example training data includes a second label.
In the network training module, each iteration of training the abnormal behavior recognition network comprises the following steps: obtaining a judgment result of each training data; the determination result is used for characterizing whether the training data comprises the specific abnormal behavior; for each similar network data, if the judgment result of the similar network data is not in accordance with the label, discarding the similar network data; obtaining the loss of the network output according to the judgment result of the training data which is not discarded and the label value; and updating the weight of the abnormal behavior recognition network according to the loss of the network output.
Therefore, similar network data which are not similar to the collected data are discarded, and the deviation of the abnormal behavior recognition network carried by the data is avoided.
In some embodiments, before the obtaining the determination result of each training data, the network training module further includes, in each iteration: acquiring action characteristics of each training data according to a backbone network; the action feature is used to determine whether the training data includes the particular abnormal behavior; in each iteration, after obtaining the loss of the network output according to the decision result and the label of the training data which are not discarded, the method further comprises: and updating the weight of the backbone network according to the loss.
Therefore, by updating the backbone network, the characteristics of the backbone network that abnormal behaviors can be extracted are better, and the accuracy of the abnormal behavior identification network is improved.
In some embodiments, in each iteration of the network training module, before the step of discarding the similar network data for each similar network data if the determination result of the similar network data does not match the label, the method further includes: inputting each training data into a discriminator; the discriminator is used for judging whether the training data is the collected data.
For each similar network data, if the determination result of the network data does not match the label, discarding the network data, including: and for each piece of training data, responding to the condition that a discriminator outputs the training data which is not the collected data and the judgment result of the training data is not consistent with the label of the training data, and discarding the training data.
Therefore, the input training data does not need to contain the characteristics of identifying whether the data is collected or not, and the complexity of the model is reduced.
According to a fifth aspect of embodiments herein, there is provided a computer-readable storage medium on which a computer program is stored, the program, when executed by a processor, implementing the above-described training data acquisition method or abnormal behavior recognition network training method.
According to a sixth aspect of embodiments herein, there is provided a computer apparatus comprising:
one or more processors;
a memory for storing one or more programs;
when the one or more programs are executed by the one or more processors, the one or more processors implement the training data acquisition method or the abnormal behavior recognition network training method described above.
The specification provides a training data acquisition method, an abnormal behavior recognition network training device, equipment and a storage medium, aiming at a specific abnormal behavior, acquiring acquisition data containing the specific abnormal behavior, acquiring network data, acquiring action characteristics of the acquisition data and the network data, determining a plurality of similar network data of the acquisition data according to the similarity of the action characteristics of each network data and the action characteristics of the acquisition data, and taking the similar network data and the acquisition data as positive sample training data of the specific abnormal behavior. Through the similarity of the action characteristics, training data which can be used as a positive sample is determined from a plurality of cheap network data, the training data acquisition efficiency is improved, and the acquisition of an abnormal behavior recognition network is further accelerated.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the specification.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present specification and together with the description, serve to explain the principles of the specification.
FIG. 1 is a flow chart illustrating a training data acquisition method according to an exemplary embodiment of the present description.
Fig. 2 is a flowchart illustrating a similar network data acquisition method according to an exemplary embodiment of the present disclosure.
Fig. 3 is a schematic diagram of similar network data shown in this specification.
FIG. 4 is a flow chart illustrating a method of abnormal behavior recognition network training according to an exemplary embodiment of the present description.
FIG. 5 is a flow chart illustrating an iterative method according to an exemplary embodiment.
FIG. 6 is a block diagram of a training data acquisition device shown in accordance with an exemplary embodiment of the present description.
FIG. 7 is a block diagram illustrating an abnormal behavior recognition network training apparatus in accordance with an exemplary embodiment of the present specification.
Fig. 8 is a hardware configuration diagram of a computer device in which a training data acquisition device or an abnormal behavior recognition network training device is located according to an exemplary embodiment.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present specification. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the specification, as detailed in the appended claims.
The terminology used in the description herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the description. As used in this specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.
It should be understood that although the terms first, second, third, etc. may be used herein to describe various information, these information should not be limited to these terms. These terms are only used to distinguish one type of information from another. For example, the first information may also be referred to as second information, and similarly, the second information may also be referred to as first information, without departing from the scope of the present specification. The word "if" as used herein may be interpreted as "at … …" or "when … …" or "in response to a determination", depending on the context.
In the management of a smart city, in order to maintain the safety and harmony of the city, it is necessary to identify abnormal behaviors (such as fighting, climbing, etc.) in the city through video data and the like. In order to identify abnormal behaviors, an abnormal behavior identification network is generally trained through a large amount of labeled data for each abnormal behavior, so as to identify the abnormal behavior through the trained abnormal behavior identification network.
However, there are some problems in this process, because training the abnormal behavior recognition network requires a large amount of labeled data, the acquisition of the data usually requires a long time, and the labeling of the data also requires a large amount of manpower. Therefore, when a training task of the abnormal behavior recognition network is received, the abnormal behavior recognition network cannot be obtained quickly due to the fact that only a small amount of labeled data exists; if the training is performed according to a small amount of data, the accuracy of the abnormal behavior recognition network is difficult to reach the available accuracy requirement.
In order to solve the above problems, the present specification provides a training data acquisition method, which acquires, for a specific abnormal behavior, acquired data including the specific abnormal behavior, acquires network data, acquires action features of the acquired data and the network data, determines similar network data of a plurality of acquired data according to a similarity between the action feature of each network data and the action feature of the acquired data, and uses the similar network data and the acquired data as positive sample training data of the specific abnormal behavior. Through the similarity of the action characteristics, training data which can be used as a positive sample is determined from a plurality of cheap network data, the training data acquisition efficiency is improved, and the acquisition of an abnormal behavior recognition network is further accelerated.
The following provides a detailed description of examples of the present specification.
In a first aspect, an embodiment of the present specification provides a training data obtaining method, which is used for obtaining, for a specific abnormal behavior, positive sample training data of the specific abnormal behavior. The method can be applied to any system device suitable for implementation, such as a server, a smart city management system or other processing devices. Fig. 1 is a flowchart illustrating a training data acquisition method according to an exemplary embodiment, and the training data acquisition method will be described in detail with reference to fig. 1.
The training data acquisition method provided by the present specification may include:
It should be noted that, in the present specification, the training data acquiring method is for one specific abnormal behavior, and if training data of a plurality of specific abnormal behaviors are to be acquired, the method may be performed in multiple times. Hereinafter, unless otherwise stated. The collected data refers to the collected data aiming at a specific abnormal behavior.
The collected data can be video or dynamic image data shot by the smart city management system, and the collected data is data containing specific abnormal behaviors. Furthermore, the collected data is already labeled data, including a first label for a particular abnormal behavior.
The network data includes: an internet published data set, and/or web crawler data, and/or data generated based on a virtual game engine. The data set disclosed by the internet refers to a training data set disclosed by the internet, in other words, a data set for some tasks, such as a K400 data set and the like, which are already finished by other internet users. The web crawler data refers to a data set searched and sorted from a network, compared with an internet public data set, the internet public data set refers to a data set which is sorted by other internet users and used for training, and the web crawler data is a data set obtained by sorting video data obtained by a plurality of crawlers. The data generated by the virtual game engine is data generated by the virtual game engine, and data not including a scene can be generated. In addition, the network data acquired in step 101 are all unlabeled data and cannot be used directly.
Since the specific abnormal behavior is generally done by people and needs at least one person to participate, each collected data or network data comprises at least one person, and data irrelevant to the specific abnormal behavior in the collected data and the network data, such as a large-area background, can be cut off through portrait identification, so that noise input into the abnormal behavior identification network is reduced.
In addition, considering that the training of the abnormal behavior recognition network cannot only have positive samples (containing data of specific abnormal behaviors) and also needs to have negative samples (not containing data of specific abnormal behaviors), the method provided by the specification is only directed to the acquisition of the training data of the positive samples. The negative sample training data is composed of two parts, one part is shot data which does not contain specific abnormal behaviors and at least comprises data of one person, and the other part is data which is obtained from the network and does not contain the specific abnormal behaviors. The specific method for acquiring the negative sample training data is described below, and is not described herein again.
Since the method provided by the present specification is used for acquiring positive sample data, and the acquired data is the positive sample data containing a tag, the first tag represents that the data carrying the tag is the data containing a specific abnormal behavior. Correspondingly, the negative examples have a second label.
And 103, acquiring each network data and the action characteristics of each acquired data.
And 105, selecting similar network data of the acquired data according to the similarity between the action characteristic of each network data and the action characteristic of the acquired data.
Next, step 103 and step 105 will be collectively described. Specifically, the actions of different abnormal behaviors are different, and the actions of different data are similar for the same abnormal behavior, so in order to obtain similar network data of collected data, similar network data needs to be found based on the action characteristics, and then the action characteristics of the collected data and the network data need to be obtained first.
The network data may be collected in advance, and corresponding action features may be extracted after collection, so the action features of the network data may be data that has been acquired and stored before a training task for training the abnormal behavior recognition network is acquired, and the action features of acquiring the network data may be action features of acquiring pre-stored network data.
For collecting data, the action features of the data need to be extracted through a backbone network (backbone), which is a network trained according to a specific data set for extracting the action features, where the specific data set refers to a data set required for training the network for extracting the action features, and such a data set is generally a data set disclosed on the network, such as a K400 data set.
One acquisition data or one network data generally corresponds to one motion feature, which is generally in the form of a vector. In other words, the input of the backbone network is a video data or a dynamic image data, and the output is a feature vector.
Under the condition that the action features of the collected data are extracted through a backbone network, the acquiring of each network data and the action features of each collected data comprises the following steps: acquiring a backbone network; extracting the action characteristic of each acquired data through the backbone network; and acquiring the action characteristics of each network data which is stored in advance and extracted through the backbone network.
The motion feature of the collected data for performing the similarity calculation may be a motion feature of any collected data, or may be a motion feature synthesized from a plurality of collected data, and the synthesized motion feature is used as the motion feature of the similarity calculation, or of course, the user may specify a motion feature of the collected data that is most representative as the motion feature of the similarity calculation, and this specification is not limited herein. A specific method for acquiring similar network data will be described in detail below, and will not be described herein again.
And 107, adding a first label to the similar network data, and using the acquired data and the selected similar network data as positive sample training data for specific abnormal behaviors.
Specifically, in order to make similar network data available for training the abnormal behavior recognition network, a first label needs to be added to the similar network data, so that supervised learning can be performed with the similar network data. By the method, positive sample training data for training the abnormal behavior recognition network aiming at the specific abnormal behavior is obtained.
Next, referring to fig. 2, a detailed description will be made on a method for acquiring similar network data, where, as shown in fig. 2, fig. 2 is a flowchart of a method for acquiring similar network data according to an embodiment of the present disclosure, and includes the following steps:
As described above, the backbone network is trained through the network public data set, and the backbone network may be trained in advance or may be trained when executed in step 201. The backbone network is used for extracting the action characteristics of video or dynamic image data.
The input of the backbone network is video or dynamic image data, and the output is a feature vector.
In step 205, the characteristics of the pre-stored network data are obtained.
The features of the network data are extracted through a pre-trained backbone network. The network data can be used for training the abnormal behavior recognition network aiming at various different abnormal behaviors, so that under the condition that the training requirement of the abnormal behavior recognition network aiming at various different abnormal behaviors exists, the action characteristics of the network data are extracted in advance, the action characteristics of the same network data do not need to be extracted repeatedly through a backbone network every time a training sample of a certain abnormal behavior is obtained, and the calculated amount can be reduced.
And step 207, synthesizing the action characteristics of all the acquired data into the central characteristics of the acquired data.
Specifically, the action features of all the collected data are normalized and averaged to be synthesized into the collected data center feature, and the collected data center feature can integrate the action features of all the collected data and can accurately reflect the features of specific abnormal behaviors.
Specifically, for each network data, feature similarity of an action feature and a collected data center feature of the network data is calculated, where the feature similarity may be cosine similarity or euclidean distance, and certainly may be other values used for representing similarity of two vectors, which is not limited herein.
The selected similar network data may be network data whose similarity with the acquired data is greater than a preset threshold. Or sorting the network data according to the similarity, and selecting the first N network data, where N may be a value preset by a user.
In addition, the number N of the selected similar network data may also be obtained according to other methods, for example, determined according to a preset number ratio of the similar network data to the collected data; in other words, the selecting the similar network data of the collected data according to the similarity between the action feature of each network data and the action feature of the collected data includes: determining the quantity N of the similar network data to be acquired according to a preset quantity proportion and the quantity of the acquired data; selecting N similar network data from the network data; the similarity between the action features of any similar network data and the action features of the collected data is not less than the similarity between the action features of any unselected network data and the action features of the collected data. The preset quantity proportion is the quantity proportion of the similar network data to be collected and the collected data.
Further, when the network data has multiple categories (for example, different categories may correspond to data crawlers from different video websites, and different categories may also correspond to different network data sets), the selected similar network data should be selected from the categories as much as possible, and specifically, after the number N of the similar network data to be selected is determined, the network data larger than a preset similarity threshold may be selected according to a preset similarity threshold, and the network data of each selected category is uniformly sampled to select the similar network data.
Further, considering that the trained abnormal behavior recognition network needs to have higher recognition capability on the collected data, and does not care whether the abnormal behavior recognition network has better recognition capability on the similar network data, therefore, in order to enable the abnormal behavior recognition network to have better recognition capability on the collected data, the number of the similar network data in the training data needs not to exceed the number of the collected data. If the similar network data is selected according to a certain proportion, the network data can be selected according to the following conditions: sampling is carried out on the collected data in a ratio of 1:2, and positive sample training data are obtained.
A schematic diagram of the selected similar network data is shown in fig. 3. In the case where there are a plurality of types of network data, when similar network data is retrieved, if it is retrieved separately for each collected data, among each type of network data, the collected data most similar to the collected data can be retrieved. In addition, some actions may carry special scenes, such as climbing and the like, and require interaction between people and things to be climbed, and since the extracted features are action features, network data (such as network data 3 in fig. 3) which does not contain the special scenes is generally retrieved, and although the network data does not contain network data of specific abnormal behaviors, for the abnormal behavior identification network, the network data does not contain the scenes, but can more clearly represent the actions, and reduce noise. In other words, although the network data does not contain specific abnormal behaviors, the characteristics of the action features can be easier to learn by the abnormal behavior recognition network, and the network data should be reserved for training.
After the positive sample training data is acquired, the negative sample training data also needs to be acquired. The specific method for acquiring the negative sample training data can be to acquire data which contains a second label and is really shot by the smart city management system and does not contain specific abnormal behaviors, the data is the same as the acquired data acquisition method, and in addition, the data often contains human actions, so that the action characteristics can be extracted. The second tag is a tag added for data that does not contain a specific abnormal behavior. And a plurality of network data with the lowest action characteristic similarity with the collected data can be obtained from the network data as negative samples.
In the second aspect, after the positive sample training data and the negative sample training data are obtained, the abnormal behavior recognition network is obtained through training by using the positive sample training data and the negative sample training data. Therefore, the present specification further provides an abnormal behavior recognition network training method, and the following will describe in detail an abnormal behavior recognition network training method provided in the present specification with reference to fig. 4.
Fig. 4 is a diagram illustrating an abnormal behavior recognition network training method according to an exemplary embodiment, which is used for training an abnormal behavior recognition network for the specific abnormal behavior based on positive sample training data acquired by the training data acquisition method of the first aspect of the present specification; the method comprises the following steps:
The positive sample training data is obtained based on the training data obtaining method, and the negative sample data obtaining method is as described above and is not described herein again. Further, to distinguish between positive and negative example training data, and to perform supervised learning, the positive example training data includes a first label and the negative example training data includes a second label.
And 403, iteratively training the abnormal behavior recognition network through the positive sample training data and the negative sample training data until the loss of the network output is less than a preset first loss threshold value, or the iteration number is greater than a preset number threshold value.
The first loss threshold and the number threshold may be selected by a user according to actual conditions, the smaller the first loss threshold is, the larger the number threshold is, the better the network training effect is, but at the same time, the calculation amount may increase, and therefore, the appropriate first loss threshold and the number threshold need to be selected in consideration of two aspects of the accuracy requirement and the calculation amount requirement of the network.
Step 403 will be described in detail below. As shown in fig. 5, fig. 5 is a flow chart illustrating an iterative method according to an exemplary embodiment of the present description, including:
And 517, updating the weight of the abnormal behavior recognition network according to the loss of the network output.
Next, steps 511 to 517 will be collectively described.
In step 511, the determination result of each training data is obtained through the input action feature of each training data, and the input action feature may be an action feature used when selecting the training data, or may be obtained again according to the updated backbone network in each iteration process. The updating of the backbone network can be to update the weight of the backbone network according to the output loss, so that the vector output by the backbone network can have better representation capability for the specific abnormal behavior, the full connection layer can better distinguish the training data containing the specific abnormal behavior from the training data not containing the specific abnormal behavior, and the training effect is better.
In other words, in each iteration, before said inputting the features of each of the training data into a full-connection layer network, the method further comprises: acquiring action characteristics of each training data according to a backbone network; the action feature is used to determine whether the training data includes the particular abnormal behavior; in each iteration, after obtaining the loss of the network output according to the decision result of the training data not discarded and the label value, the method further includes: and updating the weight of the backbone network according to the loss.
In step 511, the determination result may be obtained through a full connection layer network, and the classifier formed by the full connection layer network classifies the input motion feature vector to obtain an output determination result, and what needs to be trained in this specification is a classifier for obtaining the determination result. Through cheap data, the classifier with sufficient precision can be obtained, and thus, the classifier, namely the abnormal behavior recognition network, can be obtained quickly.
In step 513, similar network data with the determination result not matching the label is discarded, so as to prevent the final trained two classifiers from being inaccurate due to the similar network data during the next training, and in the case that the weight of the backbone network needs to be updated, discarding the similar network data can also prevent the similar network data from deviating from the representation space of the backbone network, thereby further preventing the final two classifiers from being affected.
The abnormal behavior identification network is a classifier, the corresponding labels are only the first label and the second label, the judgment result can be represented by any two numbers, the label values are represented by any two numbers, the obtained judgment result and the corresponding labels are consistent under the condition that whether the obtained judgment result and the corresponding labels contain the specific abnormal behavior or whether the obtained judgment result and the corresponding labels do not contain the specific abnormal behavior, and the judgment result and the labels are inconsistent under the other conditions. For example, regarding the determination result, 0 indicates that the training data is training data including a specific abnormal behavior, 1 indicates that the training data is training data not including a specific abnormal behavior, 0 indicates a first label, and 1 indicates a second label, if the determination result is 0, the label is 0, and according to the determination result and the meaning indicated by the first label 0, it is known that the two are matched, and if the determination result is 1, the label is 0, which indicates that the two are not matched.
In addition, it is considered that the collected data cannot be discarded because it is similar network data that needs to be discarded, and therefore it is also necessary to determine whether the data is similar network data before discarding the training data. The method for determining whether the training data is similar network data may be determined according to whether each piece of training data stored in advance is similar network data, or may be determined by training a discriminator based on the generation of a countermeasure, and determining whether the training data is network data. In other words, in each iteration, before the step of discarding the similar network data for each similar network data if the determination result of the similar network data does not match the label, the method further includes: inputting each training data into a discriminator; the discriminator is used for judging whether the training data is the collected data; for each similar network data, if the determination result of the similar network data does not conform to the label, discarding the similar network data includes: and for each piece of training data, responding to the condition that a discriminator outputs the training data which is not the collected data and the judgment result of the training data is not consistent with the label of the training data, and discarding the training data.
In step 515, the loss of the abnormal behavior recognition network may be obtained according to a loss function, where the loss function is a function preset by the user, the judgment result and the label of each training data are input, and the loss of the network is output for evaluating the quality of the abnormal behavior recognition network. The loss function may select a loss function of the two classifiers.
Because the abnormal behavior recognition network in the present specification focuses more on the recognition result of the collected data, when the calculation loss occurs, the weighting calculation may be performed, the weight of the collected data is increased, and the weight of the similar network data is decreased. If the quantity of the similar network data is less than that of the acquired training data when the training data is acquired, the acquired training data is acquired according to a certain proportion, the contribution of the acquired training data to the loss is larger than that of the similar network data, and the loss can be calculated without using a weighting calculation method.
The weight updated in step 517 is the weight of the abnormal behavior recognition network, in other words, the updated weight is the weight of the network (which may be a full connection layer) used in step 511, and the weight updating may be performed according to a gradient descent method, or according to another weight updating method, which is not limited herein.
In addition, in the above embodiment, only the abnormal behavior recognition network for a specific abnormal behavior is described, when a plurality of abnormal behavior recognition networks for different abnormal behaviors need to be trained at the same time, the plurality of networks may be trained at the same time and share the same backbone network, the weight of each abnormal behavior recognition network is updated according to the loss of each network, and the weight of the backbone network is updated according to the loss of all the abnormal behavior recognition networks. Thus, training efficiency is improved.
The foregoing description of the various embodiments is intended to highlight different aspects of the various embodiments, and all of the same or similar aspects may be referenced, and for brevity, are not repeated herein.
It will be understood by those skilled in the art that in the method of the present invention, the order of writing the steps does not imply a strict order of execution and any limitations on the implementation, and the specific order of execution of the steps should be determined by their function and possible inherent logic.
Corresponding to the embodiments of the training data acquisition method and the abnormal behavior recognition network training method, the specification also provides embodiments of a training data acquisition device, an abnormal behavior recognition network training device and a terminal applied by the training data acquisition device and the abnormal behavior recognition network training device.
As shown in fig. 6, fig. 6 is a block diagram of a training data acquisition apparatus shown in the present specification according to an exemplary embodiment, the apparatus including:
the data acquisition module 610 is configured to acquire network data and acquired data including a specific abnormal behavior.
And an action characteristic obtaining module 620, configured to obtain an action characteristic of each piece of network data and each piece of collected data.
A training data selecting module 630, configured to select similar network data of the acquired data from the network data according to a similarity between an action feature of each piece of network data and an action feature of the acquired data, and use the acquired data and the similar network data as positive sample training data for a specific abnormal behavior.
In some embodiments, the network data comprises: an internet published data set, and/or web crawler data, and/or data generated based on a virtual game engine. Such network data is more comprehensive, so that training data is more diverse.
In some embodiments, the data obtaining module 610 is specifically configured to: acquiring a backbone network; extracting the action characteristic of each acquired data through the backbone network; and acquiring the action characteristics of each network data which is stored in advance and extracted through the backbone network. Through the backbone network, the action characteristics of the collected data can be acquired, the action characteristics of the network data are extracted in advance, and when the requirements of a plurality of abnormal behavior training are met, the action characteristics of the network data do not need to be acquired for many times, so that the training efficiency is improved.
In some embodiments, the training data selecting module 630 selects similar network data of the collected data according to the similarity between the action feature of each network data and the action feature of the collected data, including: synthesizing the action characteristics of all acquired data into the characteristics of an acquired data center; and selecting similar network data of the acquired data according to the similarity of the action characteristic of each network data and the characteristic of the acquired data center. Therefore, the action characteristics of all the collected data are synthesized into the central characteristic, the central characteristic can reflect the characteristics of the abnormal behavior better, and the selected similar network data are more accurate.
In some embodiments, the training data selecting module 630 selects similar network data of the collected data according to the similarity between the action feature of each network data and the action feature of the collected data, including: determining the quantity N of the similar network data to be acquired according to a preset quantity proportion and the quantity of the acquired data; selecting N similar network data from the network data; and the similarity between the action characteristic of any one of the similar network data and the action characteristic of the collected data is not less than the similarity between the action characteristic of any one of the network data which is not selected and the action characteristic of the collected data. Therefore, the selected proportion of the similar network data to the collected data meets a certain condition, so that the abnormal behavior recognition network is not biased by the similar network data, and a better effect is achieved on the collected data.
As shown in fig. 7, fig. 7 is a block diagram of an abnormal behavior recognition network training apparatus shown in the present specification according to an exemplary embodiment, the apparatus includes:
the training data obtaining module 710 is configured to obtain training data, where the training data includes positive sample training data and negative sample training data, and the positive sample training data is obtained based on the training sample obtaining method.
And the network training module 720 is configured to iteratively train the abnormal behavior recognition network according to the positive sample training data and the negative sample training data until the loss of the network output is less than a preset first loss threshold, or the iteration number is greater than a preset number threshold.
In some embodiments, the positive example training data includes a first label and the negative example training data includes a second label. In the network training module 720, each iteration of training the abnormal behavior recognition network includes: obtaining a judgment result of each training data; the determination result is used for characterizing whether the training data comprises the specific abnormal behavior; for each similar network data, if the judgment result of the similar network data is not in accordance with the label, discarding the similar network data; obtaining the loss of the network output according to the judgment result of the training data which is not discarded and the label value; and updating the weight of the abnormal behavior recognition network according to the loss of the network output. Therefore, similar network data which are not similar to the collected data are discarded, and the deviation of the abnormal behavior recognition network carried by the data is avoided.
In some embodiments, in each iteration of the network training module 720, before the obtaining the determination result of each of the training data, the method further includes: acquiring action characteristics of each training data according to a backbone network; the action feature is used to determine whether the training data includes the particular abnormal behavior; in each iteration, after obtaining the loss of the network output according to the decision result and the label of the training data which are not discarded, the method further comprises: and updating the weight of the backbone network according to the loss. Therefore, by updating the backbone network, the characteristics of the backbone network that abnormal behaviors can be extracted are better, and the accuracy of the abnormal behavior identification network is improved.
In some embodiments, in each iteration of the network training module 720, before discarding the similar network data for each similar network data if the determination result of the similar network data does not match the label, the method further includes: inputting each training data into a discriminator; the discriminator is used for judging whether the training data is the collected data. For each similar network data, if the determination result of the network data does not match the label, discarding the network data, including: and for each piece of training data, responding to the condition that a discriminator outputs the training data which is not the collected data and the judgment result of the training data is not consistent with the label of the training data, and discarding the training data. Therefore, the input training data does not need to contain the characteristics of identifying whether the data is collected or not, and the complexity of the model is reduced.
The implementation process of the functions and actions of each module in the above device is specifically described in the implementation process of the corresponding step in the above method, and is not described herein again.
For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, wherein the modules described as separate parts may or may not be physically separate, and the parts displayed as modules may or may not be physical modules, may be located in one place, or may be distributed on a plurality of network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution in the specification. One of ordinary skill in the art can understand and implement it without inventive effort.
As shown in fig. 8, fig. 8 is a hardware structure diagram of a computer device where the above apparatus is located, and the computer device may only include a training data acquisition apparatus, may also only include an abnormal behavior recognition network training apparatus, and may also include a training data acquisition apparatus and an abnormal behavior recognition network training apparatus. The apparatus may include: a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein the processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 are communicatively coupled to each other within the device via bus 1050.
The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an Application Specific Integrated Circuit (ASIC), or one or more Integrated circuits, and is configured to execute related programs to implement the technical solutions provided in the embodiments of the present disclosure.
The Memory 1020 may be implemented in the form of a ROM (Read Only Memory), a RAM (Random Access Memory), a static storage device, a dynamic storage device, or the like. The memory 1020 may store an operating system and other application programs, and when the technical solution provided by the embodiments of the present specification is implemented by software or firmware, the relevant program codes are stored in the memory 1020 and called to be executed by the processor 1010.
The input/output interface 1030 is used for connecting an input/output module to input and output information. The i/o module may be configured as a component in a device (not shown) or may be external to the device to provide a corresponding function. The input devices may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output devices may include a display, a speaker, a vibrator, an indicator light, etc.
The communication interface 1040 is used for connecting a communication module (not shown in the drawings) to implement communication interaction between the present apparatus and other apparatuses. The communication module can realize communication in a wired mode (such as USB, network cable and the like) and also can realize communication in a wireless mode (such as mobile network, WIFI, Bluetooth and the like).
It should be noted that although the above-mentioned device only shows the processor 1010, the memory 1020, the input/output interface 1030, the communication interface 1040 and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, those skilled in the art will appreciate that the above-described apparatus may also include only those components necessary to implement the embodiments of the present description, and not necessarily all of the components shown in the figures.
Embodiments of the present specification also provide a computer-readable storage medium, on which a computer program is stored, where the computer program, when executed by a processor, implements the above-mentioned training data obtaining method or the abnormal behavior recognition network training method. The computer-readable storage medium may store only a computer program corresponding to the training data set acquisition method, or may store only a computer program corresponding to the abnormal behavior recognition network training method.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
Other embodiments of the present description will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This specification is intended to cover any variations, uses, or adaptations of the specification following, in general, the principles of the specification and including such departures from the present disclosure as come within known or customary practice within the art to which the specification pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the specification being indicated by the following claims.
It will be understood that the present description is not limited to the precise arrangements described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present description is limited only by the appended claims.
The above description is only a preferred embodiment of the present disclosure, and should not be taken as limiting the present disclosure, and any modifications, equivalents, improvements, etc. made within the spirit and principle of the present disclosure should be included in the scope of the present disclosure.
Claims (13)
1. A method of training data acquisition, the method comprising:
acquiring network data and acquisition data containing specific abnormal behaviors;
acquiring action characteristics of each network data and each acquired data;
and selecting similar network data of the acquired data from the network data according to the similarity between the action characteristic of each network data and the action characteristic of the acquired data, and taking the acquired data and the similar network data as positive sample training data aiming at specific abnormal behaviors.
2. The method of claim 1, wherein the network data comprises:
an internet published data set, and/or web crawler data, and/or data generated based on a virtual game engine.
3. The method according to any one of claims 1-2, wherein said obtaining an action characteristic of each of said network data and each of said collected data comprises:
acquiring a backbone network;
extracting the action characteristic of each acquired data through the backbone network;
and acquiring the action characteristics of each network data which is stored in advance and extracted through the backbone network.
4. The method according to any one of claims 1 to 3, wherein selecting the similar network data of the collected data according to the similarity between the action characteristic of each network data and the action characteristic of the collected data comprises:
synthesizing the action characteristics of all acquired data into the characteristics of an acquired data center;
and selecting similar network data of the acquired data according to the similarity of the action characteristic of each network data and the characteristic of the acquired data center.
5. The method according to any one of claims 1 to 4, wherein selecting the similar network data of the collected data according to the similarity between the action characteristic of each network data and the action characteristic of the collected data comprises:
determining the quantity N of the similar network data to be acquired according to a preset quantity proportion and the quantity of the acquired data;
selecting N similar network data from the network data; and the similarity between the action characteristic of any one of the similar network data and the action characteristic of the collected data is not less than the similarity between the action characteristic of any one of the network data which is not selected and the action characteristic of the collected data.
6. An abnormal behavior recognition network training method, characterized in that the method comprises:
obtaining training data, the training data comprising positive sample training data and negative sample training data, the positive sample training data obtained based on the method of any one of claims 1-5;
and iteratively training the abnormal behavior recognition network through the positive sample training data and the negative sample training data until the loss of the network output is less than a preset first loss threshold value or the iteration times are greater than a preset time threshold value.
7. The method of claim 6, wherein the positive sample training data comprises a first label and the negative sample training data comprises a second label;
each iteration of training the abnormal behavior recognition network comprises:
obtaining a judgment result of each training data; the determination result is used for characterizing whether the training data comprises the specific abnormal behavior;
for each similar network data, if the judgment result of the similar network data is not in accordance with the label, discarding the similar network data;
obtaining the loss of the network output according to the judgment result of the training data which is not discarded and the label value;
and updating the weight of the abnormal behavior recognition network according to the loss of the network output.
8. The method of claim 6 or 7, wherein, in each iteration, before the obtaining the decision result of each of the training data, the method further comprises:
acquiring action characteristics of each training data according to a backbone network; the action feature is used to determine whether the training data includes the particular abnormal behavior;
in each iteration, after obtaining the loss of the network output according to the decision result and the label of the training data which are not discarded, the method further comprises:
and updating the weight of the backbone network according to the loss.
9. The method according to any one of claims 6-8, wherein in each iteration, before said for each similar network data, if the determination result of the similar network data does not match the label, discarding the similar network data, the method further comprises:
inputting each training data into a discriminator; the discriminator is used for judging whether the training data is the collected data;
for each similar network data, if the determination result of the network data does not match the label, discarding the network data, including:
and for each piece of training data, responding to the condition that a discriminator outputs the training data which is not the collected data and the judgment result of the training data is not consistent with the label of the training data, and discarding the training data.
10. A training data acquisition apparatus, characterized in that the apparatus comprises:
the data acquisition module is used for acquiring network data and acquisition data containing specific abnormal behaviors;
the action characteristic acquisition module is used for acquiring action characteristics of each piece of network data and each piece of acquired data;
and the training data selection module is used for selecting similar network data of the acquired data from the network data according to the similarity between the action characteristic of each network data and the action characteristic of the acquired data, and taking the acquired data and the similar network data as positive sample training data aiming at a specific abnormal behavior.
11. An abnormal behavior recognition network training apparatus, characterized in that the apparatus comprises:
a training data acquisition module for acquiring training data, the training data comprising positive sample training data and negative sample training data, the positive sample training data being acquired based on the method of any one of claims 1-5;
and the network training module is used for iteratively training the abnormal behavior recognition network through the positive sample training data and the negative sample training data until the loss of the network output is less than a preset first loss threshold value or the iteration times is greater than a preset time threshold value.
12. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 9.
13. A computer device, characterized in that the computer device comprises:
one or more processors;
a memory for storing one or more programs;
when executed by the one or more processors, cause the one or more processors to implement the method of any one of claims 1-9.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111006832.0A CN113705689A (en) | 2021-08-30 | 2021-08-30 | Training data acquisition method and abnormal behavior recognition network training method |
| PCT/CN2022/077716 WO2023029397A1 (en) | 2021-08-30 | 2022-02-24 | Training data acquisition method, abnormal behavior recognition network training method and apparatus, computer device, storage medium, computer program and computer program product |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111006832.0A CN113705689A (en) | 2021-08-30 | 2021-08-30 | Training data acquisition method and abnormal behavior recognition network training method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113705689A true CN113705689A (en) | 2021-11-26 |
Family
ID=78657004
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111006832.0A Pending CN113705689A (en) | 2021-08-30 | 2021-08-30 | Training data acquisition method and abnormal behavior recognition network training method |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN113705689A (en) |
| WO (1) | WO2023029397A1 (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023029397A1 (en) * | 2021-08-30 | 2023-03-09 | 上海商汤智能科技有限公司 | Training data acquisition method, abnormal behavior recognition network training method and apparatus, computer device, storage medium, computer program and computer program product |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111368934B (en) * | 2020-03-17 | 2023-09-19 | 腾讯科技（深圳）有限公司 | Image recognition model training method, image recognition method and related device |
| CN111931713B (en) * | 2020-09-21 | 2021-01-29 | 成都睿沿科技有限公司 | Abnormal behavior detection method and device, electronic equipment and storage medium |
| CN112287857A (en) * | 2020-11-03 | 2021-01-29 | 北京享云智汇科技有限公司 | Multi-view abnormal gait recognition method |
| CN113705689A (en) * | 2021-08-30 | 2021-11-26 | 上海商汤智能科技有限公司 | Training data acquisition method and abnormal behavior recognition network training method |
-
2021
- 2021-08-30 CN CN202111006832.0A patent/CN113705689A/en active Pending
-
2022
- 2022-02-24 WO PCT/CN2022/077716 patent/WO2023029397A1/en unknown
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023029397A1 (en) * | 2021-08-30 | 2023-03-09 | 上海商汤智能科技有限公司 | Training data acquisition method, abnormal behavior recognition network training method and apparatus, computer device, storage medium, computer program and computer program product |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2023029397A1 (en) | 2023-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11483268B2 (en) | Content navigation with automated curation | |
| CN109816039B (en) | Cross-modal information retrieval method and device and storage medium | |
| CN110059468B (en) | Applet risk identification method and device | |
| CN109740018B (en) | Method and device for generating video label model | |
| CN109447156B (en) | Method and apparatus for generating a model | |
| CN109214421B (en) | Model training method and device and computer equipment | |
| CN105787133B (en) | Advertisement information filtering method and device | |
| CN110070076B (en) | Method and device for selecting training samples | |
| US11601391B2 (en) | Automated image processing and insight presentation | |
| CN114494935B (en) | Video information processing method and device, electronic equipment and medium | |
| CN113763348A (en) | Image quality determination method and device, electronic equipment and storage medium | |
| CN117409419A (en) | Image detection method, device and storage medium | |
| CN113806501B (en) | Training method of intention recognition model, intention recognition method and equipment | |
| CN113705689A (en) | Training data acquisition method and abnormal behavior recognition network training method | |
| CN114091551A (en) | Pornographic image identification method and device, electronic equipment and storage medium | |
| CN117786485A (en) | Self-training method, system, device, electronic equipment and storage medium | |
| US10789477B2 (en) | Method and apparatus for real-time detection of a scene | |
| CN110263618A (en) | The alternative manner and device of one seed nucleus body model | |
| CN115393755A (en) | Visual target tracking method, device, equipment and storage medium | |
| CN115063858A (en) | Video facial expression recognition model training method, device, equipment and storage medium | |
| CN113255766A (en) | Image classification method, device, equipment and storage medium | |
| CN111881187A (en) | Method for automatically establishing data processing model and related product | |
| CN113496243A (en) | Background music obtaining method and related product | |
| CN111723719A (en) | Video target detection method, system and device based on category external memory | |
| CN113762382B (en) | Model training and scene recognition method, device, equipment and medium |
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 | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40058659 Country of ref document: HK |