Face recognition training method, data training node, training coordination node and system

Abstract

The present disclosure provides a face recognition training method, a data training node, a training coordination node, a system, an electronic device and a computer readable storage medium to solve the problem of low efficiency and low accuracy of single data training face recognition. The method comprises: the data training node performs data training according to the face data owned by it in each round of training, and generates the corresponding training model and model information; the parameters of the training model are interacted with the training coordination node, so that the model integration is performed according to the parameters of the training model provided by all data training nodes, and the respective model adjustment information of each data training node is determined and informed; the data training node receives the corresponding model adjustment information; and the model is adjusted to update the training model. The technical scheme of the present disclosure can allow different data nodes to jointly model without sharing data, improve the face recognition training effect, and ensure that the face data is not leaked.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of face recognition, in particular to a face recognition training method, a data training node, a training coordination node, a face recognition training system, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Face recognition is a biometric technology that identifies a person based on the facial feature information of the person. It uses a camera or a camera head to collect images or video streams containing faces, and automatically detects and tracks faces in the images, and then performs a series of related technologies for face recognition. Generally, the accuracy of face recognition depends on a larger amount of data and better data quality. For machine training, data diversity is very important, but the size of the data is also important. The training and testing of a face recognition system need to be performed on millions or even tens of millions of faces. With people's concerns about privacy and monitoring, the face recognition data set of a single face recognition training party is often limited in data, so the face recognition model trained by a single data owning node cannot meet the needs of production and life. SUMMARY

[0003] In order to at least solve the technical problem that the efficiency of training a face recognition model by a single data owning node is low and the accuracy of face recognition cannot meet the requirements in the prior art, the present disclosure provides a face recognition training method, a data training node, a training coordination node, a face recognition training system, an electronic device and a computer readable storage medium, which can allow different data owners to jointly model without sharing data, improve the face recognition training effect, and ensure that face data is not leaked.

[0004] In a first aspect, the present disclosure provides a face recognition training method applied to a data training node, the method comprising:

[0005] performing data training according to the face data owned by the data training node in each round of training, and generating a training model and model information corresponding to the data training node;

[0006] interacting with the training coordination node the parameters of the training model of the data training node, so that the training coordination node integrates the training models provided by all data training nodes according to the parameters of the training models, respectively determines and informs each data training node of the model adjustment information corresponding to the data training node;

[0007] receiving the model adjustment information corresponding to the data training node informed by the training coordination node;

[0008] performing model adjustment according to the model adjustment information to update the training model of the data training node.

[0009] Further, the method further comprises:

[0010] After each round of training, the face data usage information, parameter exchange information, and training related information of each data training node are uploaded to the distributed ledger as the current round of training information to determine the contribution of each data training node to the face recognition service.

[0011] Further, the method further comprises:

[0012] After each round of training, the face data usage information, parameter exchange information, and training related information of each data training node are uploaded to the distributed ledger as the current round of training information to determine the contribution of each data training node to the face recognition service.

[0013] Further, the method further comprises:

[0014] According to the face data usage information in each round of training information, the contribution of each data training node to the face recognition service is determined.

[0015] In a second aspect, the present disclosure provides a face recognition training method applied to a training coordination node, the method comprising:

[0016] interacting with each data training node to exchange parameters of the respective training model, wherein each data training node performs data training according to the face data owned by the data training node in each round of training and generates a respective training model and model information;

[0017] integrating the training model parameters provided by all data training nodes, determining and informing each data training node of the respective model adjustment information corresponding to the data training node, so that each data training node adjusts the model according to the respective model adjustment information to update the training model.

[0018] Further, the method further comprises:

[0019] uploading the respective model adjustment information corresponding to each data training node after each round of training to the distributed ledger for storage.

[0020] Further, the method further comprises:

[0021] uploading the respective model adjustment information corresponding to each data training node after each round of training to the distributed ledger for storage.

[0022] In a third aspect, the present disclosure provides a data training node, comprising:

[0023] a training module configured to perform data training according to the face data owned by the data training node in each round of training and generate a respective training model and model information;

[0024] The first interaction module is configured to interact with the training coordination node to obtain the parameters of the training model trained by the training module, so that the training coordination node can integrate the model according to the parameters of the training model provided by all data training nodes, and determine and inform each data training node of its corresponding model adjustment information.

[0025] The receiving module is configured to receive model adjustment information corresponding to itself from the training coordination node.

[0026] The training module is also configured to adjust the model based on the model adjustment information to update its training model.

[0027] Fourthly, this disclosure provides a training coordination node, including:

[0028] The second interaction module is configured to interact with all data training nodes to exchange the parameters of their respective training models. Each data training node performs data training based on the face data it possesses in each round of training and generates its own training model and model information.

[0029] The integration module is configured to integrate the training model parameters provided by all data training nodes, determine and inform each data training node of its corresponding model adjustment information, so that each data training node can adjust its model according to its corresponding model adjustment information to update its training model.

[0030] Fifthly, this disclosure provides a face recognition training system, including a distributed ledger, a training system node as described in the third aspect, and a training coordination node as described in the fourth aspect.

[0031] In a sixth aspect, this disclosure provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes a face recognition training method as described in either the first or second aspect.

[0032] In a seventh aspect, this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the face recognition training method described in either the first or second aspect above.

[0033] Beneficial effects:

[0034] The face recognition training method, the data training node, the training coordination node, the face recognition training system, the electronic device and the computer readable storage medium provided by the present disclosure are provided. The data training node performs data training according to the face data owned by the data training node in each round of training, and generates the training model and the model information corresponding to the data training node. The data training node interacts the parameters of the training model with the training coordination node, so that the training coordination node integrates the models according to the parameters of the training model provided by all the data training nodes, respectively determines and informs each data training node of the model adjustment information corresponding to the data training node, receives the model adjustment information corresponding to the data training node informed by the training coordination node, and adjusts the model according to the model adjustment information to update the training model. The technical scheme of the present disclosure can allow different data owners to jointly model without sharing data, improve the face recognition training effect, and ensure that the face data is not leaked. The technical problems of low efficiency of training the face recognition model by a single data owning node and the face recognition accuracy cannot meet the requirements are solved. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 A flowchart of a face recognition training method provided by the first embodiment of the present disclosure is shown.

[0036] Figure 2 A flowchart of a face recognition training method provided by the second embodiment of the present disclosure is shown.

[0037] Figure 3 A face recognition training process diagram provided by the second embodiment of the present disclosure is shown.

[0038] Figure 4 An architecture diagram of a data training node provided by the third embodiment of the present disclosure is shown.

[0039] Figure 5 An architecture diagram of a training coordination node provided by the fourth embodiment of the present disclosure is shown.

[0040] Figure 6 An architecture diagram of a face recognition training system provided by the fifth embodiment of the present disclosure is shown.

[0041] Figure 7 An architecture diagram of an electronic device provided by the sixth embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0042] In order for those skilled in the art to better understand the technical solutions of the present disclosure, the present disclosure will be described in further detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments and drawings described herein are only used to explain the present disclosure, and not to limit the present disclosure.

[0043] It should be noted that the terms "first", "second", etc. in the description of the present disclosure and the claims and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence; and, without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other at will.

[0044] The terms used in the embodiments of the present disclosure are merely for the purpose of describing specific embodiments, and are not intended to limit the present disclosure. The singular forms "a", "said" and "the" used in the embodiments of the present disclosure and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise.

[0045] In the following description, the suffixes such as "module", "part" or "unit" used to represent elements are only for the convenience of description of the present disclosure, and have no specific meaning in itself. Therefore, "module", "part" or "unit" can be used mixedly.

[0046] The technical solutions of the present disclosure and how the technical solutions of the present disclosure solve the problem that the efficiency is low when a single data owner trains a face recognition model and the face recognition accuracy cannot meet the requirements will be described in detail below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments.

[0047] Figure 1 A flowchart of a face recognition training method provided by Embodiment One of the present disclosure is applied to a data training node, as shown in Figure 1 The method comprises:

[0048] Step S101: In each round of training, data training is performed according to the face data owned by it, and its corresponding training model and model information are generated;

[0049] Step S102: Interact with the training coordination node the parameters of the training model, so that the training coordination node integrates the model according to the parameters of the training model provided by all data training nodes, respectively determines and informs each data training node of the model adjustment information corresponding to each data training node;

[0050] Step S103: Receive the model adjustment information corresponding to itself informed by the training coordination node;

[0051] Step S104: Adjust the model according to the model adjustment information to update the training model.

[0052] In the embodiments of the present disclosure, the face recognition training is completed by the modeling trained by multiple data training nodes, the face information data owned by each data training node is independently saved by the data training node, in each round of training, after the data training nodes train the data respectively, each data training node generates a model and model information, the model information includes model generation time, parameters, model effect, etc., each data training node interacts the parameters of the training model with the training coordination node, based on the federated learning model training process, the training coordination node processes the relationship between each participating node, extracts the model of each data training node, integrates the model according to the parameters of the training model of all data training nodes, and informs the corresponding data training node after determining the model adjustment information corresponding to each data training node. The data training node adjusts the model according to the model adjustment information, updates the trained model, and can independently set the training round value, for example, training 500 times, or setting the model error rate to be lower than a certain value (such as 1%, 0.5% or 0.1%) to stop training; each data training node gets a consistent model, and the parameters are different face recognition models.

[0053] In the embodiments of the present disclosure, each data training node interacts with the training coordination node to exchange training model parameters, adapts data, and informs the training participating node to adjust the model by the training coordination node. The training of the participating node is independently performed, and the training coordination node adjusts according to the training process of the participating node, each data training participating node does not share, but each participating node interacts with the training coordination node to exchange training model parameters, the training coordination node integrates the model according to the training model of each data training node based on the data held by the data training node, so that each training training node adjusts the model, so that the model of each node can be applied to data of multiple sources, and the training efficiency is improved and the training effect is improved.

[0054] Further, the method further comprises:

[0055] After each round of training is completed, the face data usage information, parameter exchange information and training related information are uploaded to the distributed ledger as the current training information to determine the contribution to the face recognition service.

[0056] In each round of training, each data training node uploads information of data usage, parameter exchange information and training related information to a distributed ledger (DLT), wherein the information of data usage and the parameter exchange information include, for example, data usage time, usage quantity, training purpose, exchanged parameters, and the like, and the training related information is model information, parameter adjustment information, model effect and the like in the training process; then based on the blockchain technology, the contribution of each data training node to the face recognition service is calculated according to the data usage or training information stored in the DLT system, so as to encourage the data owner nodes to continue to contribute to the face recognition.

[0057] Further, the method further comprises:

[0058] After each round of training, the face data usage information, the parameter exchange information and the training related information are uploaded to the trusted training system for storage as the current round of training information, and part or all of the current round of training information is uploaded to the distributed ledger for storage by the trusted training system.

[0059] The information can be stored in the trusted training system according to the requirement, or can be stored in the distributed ledger. The training system and the distributed ledger are connected, and one of them can be found to find the other one. The trusted training system is used as a data transfer to solve the problem of explosion of blockchain data, and a large amount of data is stored in the trusted training system, and then a small amount of important data is stored in the DLT. The data storage can be changed according to the user's regulation. The data usage information is uploaded to the distributed ledger by the data training participant nodes and the training coordination nodes, allowing different data owners to jointly model without sharing data, and determining the workload or the amount of data provided by each data owner.

[0060] Further, the determination of the contribution to the face recognition service comprises:

[0061] The contribution to the face recognition service is calculated according to the face data usage information in each round of training information.

[0062] The training related information is uploaded to the trusted training system and the DLT system, and the contribution of each data training node to the face recognition service is calculated according to the data usage information stored in the trusted training system and the DLT system. The calculation of the contribution according to the actual usage information can ensure the fairness of the training process.

[0063] The model of each data training node can obtain the training of other data of the data, different data owners (data training nodes) are allowed to jointly model without sharing data, the face recognition training effect can be improved, and the face data is not leaked. And by linking the training participants and the blockchain, the distributed ledger only serves as a storage function, ensuring that data providers can calculate the sharing degree when using data, and encouraging data owners to continue to contribute to face recognition.

[0064] Figure 2 A flowchart of a face recognition training method provided by the second embodiment of the present disclosure is applied to a training coordination node, as shown in Figure 2 The method comprises the following steps:

[0065] Step S201: interact with all data training nodes respectively to obtain the parameters of the respective training models, wherein each data training node performs data training according to the face data owned by it in each round of training and generates a respective training model and model information;

[0066] Step S202: integrate the model according to the training model parameters provided by all data training nodes, respectively determine and inform each data training node of the respective corresponding model adjustment information, so that each data training node adjusts the model according to the respective corresponding model adjustment information to update the training model.

[0067] The training coordination node interacts with each data training node to obtain the parameters of the training model of all data training nodes, based on the federated learning model training process, the training coordination node processes the relationship between each data training node, extracts the model of each data training node, integrates the model according to the parameters of the training model of all data training nodes, and informs the corresponding data training node after determining the corresponding model adjustment information of each data training node. The model integration process can use existing model integration techniques in machine learning, such as bagging and boosting, etc. The training coordination node sends the model adjustment information to the data training node, so that it adjusts the model according to the model adjustment information.

[0068] After updating the model of the data training node, it also uploads the information of its data usage, parameter exchange information and training related information to the distributed ledger after each round of training, so as to determine its contribution to the face recognition service.

[0069] Further, the method further comprises:

[0070] The respective model adjustment information of each data training node after completing each round of training is uploaded to the distributed ledger for storage.

[0071] The model adjustment information is uploaded to the distributed ledger, the training coordination node knows the amount of data of all data training nodes participating in the training, and the workload of the corresponding model adjustment information is also recorded in the distributed ledger, and the contribution of the training coordination node to the face recognition training can be determined accordingly.

[0072] Further, the method further comprises:

[0073] The corresponding model adjustment information of each data training node after completing each round of training is uploaded to the trusted training system for storage, and then uploaded to the distributed ledger for storage by the trusted training system.

[0074] The training coordination node can upload the model information and the corresponding adjustment information to the distributed ledger or to the trusted training system, and upload the model information and the corresponding adjustment information to the distributed ledger by the trusted training system, and calculate the contribution degree of the training coordination node according to the model adjustment information in the trusted training system or the distributed ledger.

[0075] In order to more clearly describe the technical solutions of the present disclosure, please refer to Figure 3 The training parties include participants (i.e. data training nodes), training coordination nodes, distributed ledgers and trusted training systems; the face recognition training process is as follows: there are three face data owners and participants (i.e. data training nodes) A, B and C in the face recognition training; in each round of training, the three data owners and participants A, B and C train data respectively, each participant generates a model and obtains model information (including generation time, parameters, model effect, etc.), each participant interacts with the training coordination node to train the model parameters, and performs data adaptation, and the training coordination node informs the training participants to adjust the model; in each round of training, the data owners and participants can upload evidence information (including use time, use quantity, training purpose, exchanged parameters, etc.) of data use and parameter exchange to the distributed ledger (DLT) and the trusted training system; the training participants and the coordination node can also directly upload the training related information to the DLT system, or upload the training related information to the DLT through the trusted training system; through the trusted training system, any training participant can calculate the contribution degree of each participant to the face recognition service according to the data use information stored in the trusted training system and the DLT system.

[0076] Figure 4 An architecture diagram of a data training node provided in Embodiment Three of the present disclosure is shown in FIG. 3, which comprises: Figure 4

[0077] A training module 11 is configured to perform data training according to the face data owned by the data training node in each round of training, and generate the corresponding training model and model information;

[0078] ​a first interaction module 12 configured to interact with the training coordination node to exchange parameters of the trained model of the training module, so that the training coordination node integrates the model according to the parameters of the trained model provided by all data training nodes, and determines and informs each data training node of the corresponding model adjustment information of the data training node respectively;

[0079] a receiving module 13 configured to receive the corresponding model adjustment information informed by the training coordination node;

[0080] The training module 11 is further configured to adjust the model according to the model adjustment information to update the trained model.

[0081] Further, the data training node further comprises a first uploading module 14;

[0082] The first uploading module is configured to upload the face data usage information, parameter exchange information and training related information of the data training node as the current training information to the distributed ledger for storage after each round of training of the training module 11 is completed, so as to determine the contribution degree of the data training node to the face recognition service.

[0083] Further, the data training node further comprises a first uploading module 14;

[0084] The first uploading module 14 is further configured to upload the face data usage information, parameter exchange information and training related information of the data training node as the current training information to the trusted training system for storage after each round of training of the training module 11 is completed, and further upload part or all of the current training information to the distributed ledger for storage through the trusted training system.

[0085] Further, the determination of the contribution degree of the data training node to the face recognition service comprises:

[0086] According to the face data usage information in each round of training information, the contribution degree of the data training node to the face recognition service is calculated.

[0087] Figure 5 An architecture diagram of a training coordination node provided in Embodiment Four of the present disclosure is shown in Figure 5 The architecture diagram of the training coordination node comprises:

[0088] a second interaction module 21 configured to interact with all data training nodes respectively to exchange parameters of the trained model of each data training node, wherein each data training node generates a trained model and model information by performing data training according to the face data owned by the data training node in each round of training;

[0089] Integration module 22 is configured to integrate the training model parameters provided by all data training nodes, determine and inform each data training node of its corresponding model adjustment information, so that each data training node can adjust its model according to its corresponding model adjustment information to update its training model.

[0090] Furthermore, the training coordination node also includes a second upload module 23;

[0091] The second upload module 23 is configured to upload the corresponding model adjustment information after each data training node completes each round of training to the distributed ledger for storage.

[0092] Furthermore, the training coordination node also includes a second upload module 23;

[0093] The second upload module 23 is configured to upload the model adjustment information corresponding to each data training node after each round of training to the trusted training system for storage, and then upload it to the distributed ledger for storage through the trusted training system.

[0094] Figure 6 This is an architecture diagram of a face recognition training system provided in Embodiment 5 of this disclosure, as follows: Figure 6 As shown, it includes a distributed ledger 3, a data training node 1 as described in any of Embodiment 3, and a training coordination node 2 as described in any of Embodiment 4.

[0095] The data training node, training coordination node, and face recognition training system of this disclosure are used to implement the face recognition training methods in Method Embodiment 1 and Method Embodiment 2, so the description is relatively simple. For details, please refer to the relevant descriptions in Method Embodiment 1 and Method Embodiment 2 above, which will not be repeated here.

[0096] In addition, such as Figure 7 As shown, Embodiment Six of this disclosure also provides an electronic device, including a memory 10 and a processor 20. The memory 10 stores a computer program. When the processor 20 runs the computer program stored in the memory 10, the processor 20 executes the various possible methods described above.

[0097] The memory 10 is connected to the processor 20. The memory 10 can be a flash memory, a read-only memory, or another type of memory. The processor 20 can be a central processing unit or a microcontroller.

[0098] Furthermore, embodiments of this disclosure also provide a computer-readable storage medium storing a computer program, which is executed by a processor using the various possible methods described above.

[0099] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, computer program modules or other data. The computer-readable storage medium includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile discs (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium which the node can use to store the desired information and which can be accessed by the computer.

[0100] It can be understood that the above embodiments are only exemplary embodiments adopted for illustrating the principles of the present disclosure, and the present disclosure is not limited thereto. Various modifications and improvements can be made by those of ordinary skill in the art without departing from the spirit and essence of the present disclosure, and these modifications and improvements are also considered to be within the protection scope of the present disclosure.

Claims

1. A face recognition training method, characterized in that, Applied to data training nodes, the method includes: In each round of training, data training is performed based on the facial data it possesses, and its corresponding training model and model information are generated; It interacts with the training coordinating node to exchange the parameters of its training model, so that the training coordinating node can integrate the model based on the parameters of the training model provided by all data training nodes, and determine and inform each data training node of its corresponding model adjustment information. Receive model adjustment information corresponding to itself from the training coordination node; The model is adjusted based on the model adjustment information to update its training model.

2. The method according to claim 1, characterized in that, The method further includes: After each round of training, the face data usage information, parameter exchange information, and training-related information are uploaded to the distributed ledger as training information for that round to determine their contribution to the face recognition service.

3. The method according to claim 1, characterized in that, The method further includes: After each round of training is completed, the face data usage information, parameter exchange information, and training-related information are uploaded to the trusted training system for storage as training information for this round. Then, the trusted training system uploads part or all of the training information for this round to the distributed ledger for storage.

4. The method according to claim 2, characterized in that, The determination of its contribution to the facial recognition service includes: The contribution of the face data in each round of training information to the face recognition service is calculated.

5. A face recognition training method, characterized in that, Applied to training coordinating nodes, the method includes: Each training node interacts with all data training nodes to obtain the parameters of its own training model. In each round of training, each data training node trains its own training model and generates its own training model and model information based on the face data it possesses. The model is integrated based on the training model parameters provided by all data training nodes. The corresponding model adjustment information is determined and communicated to each data training node so that each data training node can adjust its model according to its corresponding model adjustment information to update its training model.

6. The method according to claim 5, characterized in that, The method further includes: After each training round, the model adjustment information corresponding to each data training node is uploaded to the distributed ledger for storage.

7. The method according to claim 5, characterized in that, The method further includes: After each training round, the model adjustment information corresponding to each data training node is uploaded to the trusted training system for storage, and then uploaded to the distributed ledger for storage through the trusted training system.

8. A data training node, characterized in that, include: The training module is configured to perform data training based on the face data possessed by the data training node in each round of training, and generate its corresponding training model and model information. The first interaction module is configured to interact with the training coordination node to obtain the parameters of the training model trained by the training module, so that the training coordination node can integrate the model according to the parameters of the training model provided by all data training nodes, and determine and inform each data training node of its corresponding model adjustment information. The receiving module is configured to receive model adjustment information corresponding to itself from the training coordination node. The training module is also configured to adjust the model based on the model adjustment information to update its training model.

9. A training coordination node, characterized in that, include: The second interaction module is configured to interact with all data training nodes to exchange the parameters of their respective training models. Each data training node performs data training based on the face data it possesses in each round of training and generates its own training model and model information. The integration module is configured to integrate the training model parameters provided by all data training nodes, determine and inform each data training node of its corresponding model adjustment information, so that each data training node can adjust its model according to its corresponding model adjustment information to update its training model.

10. A face recognition training system, characterized in that, It includes a distributed ledger, a training system node as described in claim 8, and a training coordination node as described in claim 9.

11. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor performs the face recognition training method as described in any one of claims 1-4 or 5-7.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the face recognition training method as described in any one of claims 1-4 or 5-7.

Images