CN114549983A

CN114549983A - Computer vision model training method and device, electronic equipment and storage medium

Info

Publication number: CN114549983A
Application number: CN202210152495.4A
Authority: CN
Inventors: 窦浩轩; 甘伟豪
Original assignee: Beijing Sensetime Technology Development Co Ltd
Current assignee: Beijing Sensetime Technology Development Co Ltd
Priority date: 2022-02-18
Filing date: 2022-02-18
Publication date: 2022-05-27

Abstract

The present disclosure relates to a computer vision model training method and apparatus, an electronic device, and a storage medium, in which a plurality of sets of model parameters from a plurality of electronic devices are acquired to determine a vision model corresponding to each set of model parameters according to a preset model frame, and a plurality of synthetic images are generated according to each vision model. Further, model distillation is carried out on each visual model through each synthetic image to obtain a target visual model. According to the embodiment of the disclosure, a plurality of visual models can be obtained by performing model parameter communication with a plurality of electronic devices once, and model distillation is performed through a synthetic image under the condition that training sets of the electronic devices are not received, so that a target visual model fusing characteristics of the plurality of visual models is obtained. The training method disclosed by the embodiment of the disclosure can fuse a plurality of visual model characteristics in a model distillation mode, and effectively protect the privacy of each electronic device while fusing the model characteristics.

Description

Computer vision model training method and device, electronic equipment and storage medium

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a computer vision model training method and apparatus, an electronic device, and a storage medium.

Background

At present, in the technical field of computers, in order to avoid data privacy disclosure, a data generation end is generally bound with a model training end. However, the training mode protects data privacy and enables each terminal to form a data island, data of different terminals cannot be gathered together for training, and the maximum potential is difficult to be exerted.

Disclosure of Invention

The disclosure provides a computer vision model training method and device, electronic equipment and a storage medium technical scheme, and aims to protect privacy and fuse the characteristics of models in terminals.

According to a first aspect of the present disclosure, there is provided a computer vision model training method, comprising:

acquiring a plurality of visual models;

generating a plurality of synthetic images according to the visual models respectively;

and performing model distillation by taking each synthetic image as the input of each visual model to obtain a target visual model.

In one possible implementation, the generating a plurality of composite images according to the respective visual models includes:

determining an initial plurality of noise images;

inputting each initial noise image into each visual model respectively for iterative updating until the value of the first loss function meets the convergence condition to obtain a composite image corresponding to each noise image,

wherein a value of the first loss function is determined based on the first loss, the second loss, and the third loss.

In one possible implementation, the first loss is used to characterize the loss of the corresponding visual model in detecting the noisy image;

the second loss is used for representing the similarity degree of the noise image and a real image;

the third loss is used for representing the loss of the noise image generated in the process of transmitting each batch processing layer of the corresponding visual model.

In one possible implementation manner, the determining of the first loss includes:

determining an initial first labeling result corresponding to each noise image;

inputting each initial noise image into each visual model respectively to obtain a corresponding second labeling result;

and determining a first loss according to the first labeling result and the second labeling result of each noise image.

In one possible implementation, the second loss is determined from pixel values in the noise image.

In a possible implementation manner, the third loss is determined according to the first feature map output by each batch processing layer after the noise image is input into the corresponding visual model, and the second feature map input by each batch processing layer.

In a possible implementation manner, the performing model distillation by using each of the synthesized images as an input of each of the visual models to obtain a target visual model includes:

determining a first visual model for transferring detection performance of other visual models in each visual model, and taking other visual models except the first visual model as second visual models;

and performing model distillation by taking each synthetic image as the input of the first visual model and each second visual model respectively to obtain a target visual model.

In a possible implementation manner, the performing model distillation by using each of the synthesized images as an input of the first visual model and each of the second visual models to obtain a target visual model includes:

inputting each synthesized image into a first visual model to obtain a first detection result;

inputting each composite image into a second visual model respectively to obtain a second detection result;

and iteratively training the first visual model through a second loss function to obtain a target visual model, wherein the second loss function is determined according to the first detection result and each second detection result.

In one possible implementation, the method further includes:

in response to receiving a model update request, returning target model parameters for the target visual model.

In one possible implementation, the second loss function is a sum of the norm of L2 of the first detection result and each of the second detection results.

In one possible implementation, the method further includes:

acquiring real images stored in the plurality of electronic devices;

the step of performing model distillation by taking each synthetic image as the input of the first visual model and each second visual model respectively to obtain a target visual model comprises the following steps:

and performing model distillation by taking each real image and each synthetic image as the input of the first visual model and each second visual model respectively to obtain a target visual model.

According to a second aspect of the present disclosure, there is provided a computer vision model training apparatus comprising:

a model determination module for obtaining a plurality of visual models;

an image generation module for generating a plurality of synthetic images according to the visual models, respectively;

and the model distillation module is used for performing model distillation by taking each synthetic image as the input of each visual model respectively to obtain a target visual model.

In one possible implementation, the image generation module includes:

an image initialization sub-module for determining an initial plurality of noise images;

an image iteration submodule, configured to input each initial noise image into each visual model respectively for iterative update until a value of the first loss function satisfies a convergence condition, to obtain a composite image corresponding to each noise image,

determining an initial first labeling result corresponding to each noise image;

In a possible implementation manner, the third loss is determined according to the first feature map output by each batch layer after the noise image is input into the corresponding visual model, and the second feature map input by each batch layer.

In one possible implementation, the model distillation module includes:

the model selection submodule is used for determining a first visual model for transferring the detection performance of other visual models in each visual model and taking other models except the first visual model as a second visual model;

and the model distillation submodule is used for performing model distillation by taking each synthetic image as the input of the first visual model and each second visual model respectively to obtain a target visual model.

In one possible implementation, the model distillation submodule includes:

the first detection unit is used for inputting each composite image into a first visual model to obtain a first detection result;

the second detection unit is used for inputting each composite image into a second visual model respectively to obtain a second detection result;

and the iterative training unit is used for iteratively training the first visual model through a second loss function to obtain a target visual model, and the second loss function is determined according to the first detection result and each second detection result.

In one possible implementation, the apparatus further includes:

and the parameter sending module is used for responding to the received model updating request and returning the target model parameters of the target visual model.

In one possible implementation, the apparatus further includes:

a real image acquisition module for acquiring real images stored in the plurality of electronic devices;

the model distillation submodule includes:

and the model distillation unit is used for performing model distillation by taking each real image and each synthetic image as the input of the first visual model and each second visual model respectively to obtain a target visual model.

According to a third aspect of the present disclosure, there is provided an electronic device comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the memory-stored instructions to perform the above-described method.

According to a fourth aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the above-described method.

In the embodiment of the disclosure, a plurality of visual model characteristics can be fused in a model distillation mode, and the privacy of each electronic device is effectively protected while the model characteristics are fused.

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 disclosure. Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure.

FIG. 1 illustrates a system diagram of a computer vision model training method applying embodiments of the present disclosure;

FIG. 2 illustrates a flow chart of a method of computer vision model training in accordance with an embodiment of the present disclosure;

FIG. 3 shows a flow diagram of a process of determining a composite image according to an embodiment of the disclosure;

FIG. 4 is a schematic diagram illustrating one type of determining a composite image according to an exemplary embodiment;

FIG. 5 is a schematic diagram illustrating another determination of a composite image in accordance with an exemplary embodiment;

FIG. 6 shows a schematic diagram of iteratively generating a composite image, according to an embodiment of the present disclosure;

FIG. 7 shows a schematic diagram of a model distillation process according to an embodiment of the present disclosure;

FIG. 8 is a schematic diagram illustrating a parameter communication process in accordance with an exemplary embodiment;

FIG. 9 is a schematic diagram illustrating another parameter communication process in accordance with an exemplary embodiment;

FIG. 10 shows a schematic diagram of a computer vision model training apparatus in accordance with an embodiment of the present disclosure;

FIG. 11 is a block diagram of an electronic device shown in accordance with an exemplary embodiment;

FIG. 12 is a block diagram illustrating another electronic device in accordance with an exemplary embodiment.

Detailed Description

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

The term "and/or" herein is merely an association describing an associated object, meaning that three relationships may exist, e.g., a and/or B, may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B, C, and may mean including any one or more elements selected from the group consisting of A, B and C.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.

Fig. 1 shows a system diagram of a computer vision model training method to which an embodiment of the present disclosure is applied. As shown in fig. 1, in one possible implementation manner, a system for implementing the computer vision model training method of the embodiment of the present disclosure includes a first electronic device 10 and a plurality of second electronic devices 11 connected to the first electronic device 10 through a network. Wherein, each second electronic device 11 has a visual model deployed therein, so as to train the visual model through the image data acquired by the visual model. Optionally, the framework of the visual model in each second electronic device 11 is the same, and the model parameters differ only due to the difference in training sets. The first electronic device 10 receives the model parameters obtained by the second electronic devices 11 through training corresponding to the training set, determines the visual models in the second electronic devices 11 in the first electronic device 10, and fuses the visual models in the second electronic devices 11 by generating synthetic images and performing model distillation to obtain target visual models with the characteristics of the visual models.

Alternatively, the training set in each second electronic device 11 may be determined by receiving images acquired by a particular electronic device, or by the second electronic device 11 acquiring images directly.

Further, the first electronic device 10 executing the computer vision model training method in the system and the second electronic device 11 providing the model parameters may be both terminal devices or electronic devices such as servers. The terminal device may be a User Equipment (UE), a mobile device, a User terminal, a cellular phone, a cordless phone, a Personal Digital Assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, or other devices capable of performing data processing, and the detection method may be implemented by a processor calling a computer-readable instruction stored in a memory. Alternatively, when the electronic device is a server, the computer vision model training method may be executed by the server. Alternatively, the server may be a single server or a cluster of multiple servers.

The embodiment of the disclosure can be applied to any scene where a plurality of computer vision models are fused, for example, feature fusion is performed on face recognition models in different smart phones to improve the universality of the face recognition models. Or the image classification models in different terminal devices are subjected to feature fusion so as to improve the accuracy of the classification result of the image classification model.

FIG. 2 shows a flow diagram of a computer vision model training method in accordance with an embodiment of the present disclosure. As shown in fig. 2, when performing computer vision model training, the embodiment of the present disclosure may include the following steps:

and step S10, acquiring a plurality of visual models.

In one possible implementation, a plurality of visual models are deployed in one or more electronic devices, and each visual model is trained based on a training set in the corresponding electronic device. The training set in each electronic device may be determined by receiving a plurality of images acquired by a specific image acquisition device, or may be determined directly by a plurality of images acquired by an image acquisition unit built in the electronic device. The electronic device executing the computer vision model training method of the embodiment of the disclosure can obtain a plurality of vision models by obtaining a plurality of sets of model parameters from a plurality of electronic devices respectively and determining the vision model corresponding to each set of model parameters according to a preset model frame.

Because the training sets in the electronic devices are different and not communicated with each other, the model parameters corresponding to the trained visual models are also different. Furthermore, a plurality of visual models can be determined by obtaining a plurality of sets of model parameters corresponding to the visual models in the plurality of electronic devices, so as to further fuse the characteristics of the visual models. Optionally, each set of model parameters includes all parameters of the visual model in the corresponding electronic device.

In a possible implementation manner, the manner of obtaining the model parameters transmitted by the electronic device may be to send a parameter obtaining request to each electronic device, and the electronic device returns the corresponding model parameters after receiving the parameter obtaining request. Or, each electronic device may also automatically upload the model parameters after the corresponding visual model training is completed.

Further, the structures of the visual models in the electronic devices may be the same or different. Under the condition that the structures of the visual models are the same, only the model parameters of the visual models in the electronic equipment need to be acquired. Under the condition that the structures of the visual models are different, the model parameters of the visual models can be acquired, and meanwhile, the corresponding model frames can be acquired.

In a possible implementation manner, a model frame having the same structure as the visual model in each electronic device is preset, and then the corresponding visual model is generated in the electronic device receiving the model parameters according to each set of model parameters. And obtaining the visual model obtained by training each electronic device under the condition of only obtaining the model parameters by the mode.

Further, when the visual model structures deployed in the electronic devices are different, the electronic device receiving the model parameters receives multiple sets of model parameters transmitted by the electronic devices, and also receives the model frames corresponding to the sets of model parameters. And obtaining the visual model obtained by training each electronic device according to each group of model parameters and the corresponding model framework.

In step S20, a plurality of composite images are generated from each of the visual models.

In one possible implementation, after determining the plurality of visual models, a plurality of composite images are generated based on each visual model. The determination process of the synthetic images can further carry out model distillation according to the synthetic images, and fuse the performance of each visual model under the condition of not receiving the training set corresponding to each visual model.

FIG. 3 shows a flow diagram of a process of determining a composite image according to an embodiment of the disclosure. As shown in fig. 3, in one possible implementation, the process of determining a composite image according to a visual model according to an embodiment of the present disclosure may include the following steps:

step S21, an initial plurality of noise images is determined.

In one possible implementation, the initial plurality of noise images may be obtained by way of random initialization by the electronic device. For example, a plurality of noise images may be initialized by randomly adding noise to a preset blank image. Alternatively, the noise image may be generated by gaussian noise, salt and pepper noise, or the like. Optionally, while determining the initial plurality of noise images, the first labeling result corresponding to each noise image may also be randomly generated according to the output result of each visual model.

Further, one or more data may be included in the first annotation result. For example, when the output of the visual model is the detection frame coordinates, the detection frame coordinates corresponding to each noise image are randomly determined as the first labeling result. And when the output of the visual model is the image category, randomly determining the corresponding category of each noise image as a first labeling result. And when the output result of the visual model is the detection frame coordinate and the image category, randomly determining the detection frame coordinate and the detection frame category corresponding to each noise image as a first labeling result.

Step S22, inputting each initial noise image into each visual model, and performing iterative update until the value of the first loss function satisfies a convergence condition, to obtain a composite image corresponding to each noise image.

In one possible implementation, a plurality of initially obtained initial noise images determined to be initially obtained are input into each visual model, and the predicted second labeling result is output. And the noise images can be input into the visual models to obtain a plurality of second labeling results output by the visual models. Or, each noise image is only input into one visual model, and a second labeling result output by the corresponding visual model is obtained. For example, when the visual model is used for image classification, the output second labeling result may be the detection frame coordinates and/or the image category obtained by detecting the noise image.

In one possible implementation, for each noise image, the visual model input by the noise image is iteratively updated until the value of the first loss function satisfies the convergence condition, and a composite image corresponding to the noise image and the visual model is obtained. Wherein the value of the first loss function is determined based on the first loss, the second loss, and the third loss for each noise image. The first loss represents the loss of the corresponding visual model in the process of detecting the noise image, the second loss represents the similarity degree of the noise image and the real image, and the third loss represents the loss of the noise image in the process of transmitting each batch processing layer of the corresponding visual model.

In one possible implementation, the first loss function may include a first loss, a second loss, and a third loss. Alternatively, the first loss function may be a weighted sum of the first loss, the second loss, and the third loss. Wherein the first loss is used to characterize the loss generated by the corresponding visual model in detecting the noisy image. The second loss is used for representing the similarity degree of the noise image and the real image, and the third loss is used for representing the loss of the noise image generated in the process of transferring each batch processing layer of the corresponding visual model. Further, the determining process of the first loss includes determining a first labeling result corresponding to each initial noise image, and inputting each initial noise image into each visual model respectively to obtain a corresponding second labeling result. And determining a first loss according to the first labeling result and the second labeling result of each noise image. And initializing the first labeling result and the initial noise image together. The second penalty is determined from the values of the pixels in the noisy image. And determining the third loss according to the first characteristic diagram output by each batch processing layer after the noise image is input into the corresponding visual model and the second characteristic diagram input into each batch processing layer.

The embodiment of the disclosure determines the first loss function based on the pixel degrees of the noise image and the real image, the detection loss of the input visual model and the difference of input and output of each layer in the input visual model, and can iterate the noise image based on three different losses. The synthetic image obtained through the first loss function can represent the characteristics of the input visual model corresponding to the training set, and the accuracy of the subsequent model distillation result is improved.

Alternatively, the first loss may be determined according to the content of the output result of each visual model, and represents the difference between the detection result of the visual model and the actual result, that is, the first labeling result corresponding to the noise image input to the visual model, and the second labeling result output by the visual model. For example, when the content included in the first annotation result and the second annotation result is of an image category, the first loss can be determined by calculating the cross entropy. When the contents included in the first and second labeling results are the coordinates of the detection frame, the first loss may be determined by calculating the norm of L1. Further, when the first annotation result and the second annotation result include a plurality of contents, the first loss manner may be obtained by calculating a weighted sum of the contents.

It is easily understood that the above-mentioned manner of determining the difference between the model detection result and the actual result is an alternative exemplary manner in the embodiments of the present disclosure. In addition to the above manners, the first loss may be determined by a plurality of different loss calculation manners based on the difference of the visual model output in the application process, which is not described herein again.

Further, the second loss represents the pixel degree of the noise image and the real image, and can be determined by the pixel value of each pixel position of the noise image. Wherein a lower value of the second loss indicates that the noisy image is more similar to the real image. For example, for each noise image, the pixel value at each pixel position may be obtained first, and the second loss may be determined by calculating the L2 norm of each pixel value, or the total variation (total variation) of each pixel value. Alternatively, the second penalty may also be determined by calculating a weighted sum of the L2 norm and the total variation for each pixel value. The total variation represents the sum of the signal variation degrees, and can be determined by calculating the square of the difference between each pixel value and each adjacent pixel and then calculating the sum.

In a possible implementation manner, each visual model includes a plurality of batch processing layers, and each noise image is sequentially processed through the plurality of batch processing layers after being input into the visual model, so as to obtain a final output result. The input and the output of each batch processing layer are both a characteristic image, and the characteristic image input and the characteristic image output by each batch processing layer have certain difference. Alternatively, the third loss may characterize the batch layer input-output differences in the visual model. For example, for each visual model, the mean and standard deviation of the input feature images and the mean and standard deviation of the output feature images for each batch layer may be determined. The method comprises the steps of firstly calculating the L2 norm sum between the input characteristic image mean value and the output characteristic image mean value of each batch processing layer, and calculating the L2 norm sum between the input characteristic image standard deviation and the output characteristic image standard deviation, and then calculating the weighted sum of the two L2 norm sums to obtain a third loss.

In one possible implementation, after obtaining the first loss function of each noise image input to the visual model, each noise image is iterated through a gradient descent method, and the iteration is stopped until the value of the first loss function meets a preset convergence condition to obtain a composite image. The convergence condition may be that a value of the first loss function is smaller than a convergence threshold, or that the value of the first loss function is a minimum value in a preset number of iterations.

FIG. 4 is a schematic diagram illustrating one type of determining a composite image according to an exemplary embodiment. As shown in fig. 4, in an alternative implementation, for each of the noise images 40 generated randomly, a different plurality of visual models 41 may be input for iteration. Due to the different model parameters of the visual models 41, the same noise image 40 may obtain different N composite images 42 after N iterations of the visual models 41.

FIG. 5 is a schematic diagram illustrating another determination of a composite image according to an exemplary embodiment. As shown in fig. 5, in an alternative implementation, for each noise image 50 generated randomly, each noise image may be input to only one visual model 51 for iteration to obtain a corresponding composite image 52. The generation mode of the synthetic image can increase the difference of different synthetic images and further improve the accuracy of the model distillation result.

FIG. 6 shows a schematic diagram of iteratively generating a composite image, according to an embodiment of the present disclosure. As shown in fig. 6, in a possible implementation manner, the composite image generation method according to the embodiment of the disclosure is to determine a noise image 60, input the noise image into the visual model 61, and calculate a first loss function 62 according to the second labeling result output by the visual model 61, the pixel values of the pixel positions of the noise image 60, and the input and output of each batch layer in the visual model 61. Further, it is determined whether the value of the first penalty function 62 satisfies a preset convergence condition 63, and if so, a composite image 64 corresponding to the noise image 60 is obtained, and if not, the noise image 61 is adjusted, and the next iteration is performed.

Each synthetic image generated by the embodiment of the disclosure is generated iteratively based on each visual model, and bears the characteristics of each visual model, so that each model can still be fused through model distillation without a real image.

And step S30, performing model distillation by taking each synthetic image as the input of each visual model to obtain a target visual model.

In one possible implementation, a first visual model for migrating detection performance of other visual models may be determined among the visual models, and the other visual models except the first visual model may be used as a second visual model. And then, taking each synthetic image as the input of the first visual model and each second visual model respectively to carry out model distillation to obtain the target visual model. The distillation process may include inputting each synthetic image into the first visual model to obtain a first detection result, inputting each synthetic image into the second visual model to obtain a second detection result, iteratively training the first visual model through a second loss function to obtain a target visual model, and determining the second loss function according to the first detection result and each second detection result.

Optionally, the first test result and the second test result content are also determined from the output of the visual model, including one or more data. The second loss function may be a sum of the norm of L2 of the first detection result and each of the second detection results. That is, the second loss function may be determined by calculating the L2 norm sums of the same data in the first detection result and each second detection result, respectively. For example, when the detection frame coordinates are included in both the first detection result and the second detection result, the sum of the norms L2 of the detection frame coordinates in the first detection result and the detection frame coordinates in each of the second detection results is calculated to obtain the second loss function. Or, when the first detection result and the second detection result both include the detection frame coordinates and the image type, calculating the sum of the L2 norms of the detection frame coordinates in the first detection result and the detection frame coordinates in each second detection result, and the sum of the L2 norms of the image type in the first detection result and the image type in each second detection result, and calculating the weighted sum of the two L2 norms to obtain the second loss function.

Further, the process of performing model distillation through the second loss function may be adjusting the second loss function through a gradient descent method until the value of the second loss function satisfies a preset condition, so as to obtain the target visual model.

Fig. 7 shows a schematic diagram of a model distillation process according to an embodiment of the present disclosure. As shown in fig. 7, in one possible implementation, each composite image 70 is input into a first visual model 71 and each second visual model 72, the first visual model 71 detects the input composite image 70 to obtain a first detection result 73, and each second visual model 72 detects the input composite image 70 to obtain a second detection result 74. Further, a second loss function 75 is obtained by calculating the first detection result 73 and each second detection result 74, model distillation is performed by iterating the second loss function 75, and model characteristics of each second visual model 72 are fused to the first visual model 71, thereby obtaining a target visual model having all the visual model characteristics.

In an optional implementation manner of the embodiment of the present disclosure, in order to improve the authenticity of the final distillation result, a training set may be further constructed by acquiring, by a receiving device of each set of model parameters, a part of the real image together with the synthetic image, so as to perform model distillation training. That is, the electronic device may further acquire real images stored in a plurality of electronic devices, and perform model distillation using each of the real images and each of the composite images as input of the first visual model and each of the second visual models, respectively, to obtain the target visual model.

Fig. 8 is a schematic diagram illustrating a parameter communication procedure in accordance with an example embodiment. As shown in fig. 8, in one possible implementation, the first electronic device 81 receives model parameters transmitted by the plurality of second electronic devices 80, and performs model distillation on the visual model in each second electronic device 80 to obtain a target visual model having all the characteristics of the visual model. Further, in order to improve the performance of the visual model in each second electronic device 80, after the first electronic device 81 processes the target visual model to obtain a target visual model, each target model parameter in the target visual model is returned to each second electronic device 80, and the second electronic device 80 updates the visual model according to the target model parameter.

Fig. 9 is a schematic diagram illustrating another parameter communication procedure in accordance with an exemplary embodiment. As shown in FIG. 9, in one possible implementation, target model parameters of the target visual model may be returned in response to receiving a model update request.

The first electronic device 91 receives the model parameters sent by the second electronic devices 90, and performs model distillation on the visual model in each second electronic device 90 to obtain a target visual model with all visual model characteristics. Further, when the second electronic device 90 needs to update its deployed visual model, it will send a model update request to the first electronic device 91, and the first electronic device 91 returns the target model parameters of the target visual model to the second electronic device 90 that sent the model update request for model update.

The embodiment of the disclosure can obtain a plurality of visual models trained on each electronic device by only communicating model parameters with a plurality of electronic devices once. Meanwhile, under the condition of not receiving training sets of the electronic equipment, a synthetic image is generated through the visual models, a training set for model distillation is constructed, and a target visual model fusing the characteristics of the visual models is obtained. The training method provided by the embodiment of the disclosure can fuse a plurality of visual model characteristics in a model distillation mode, and effectively protect the privacy of each electronic device while fusing the model characteristics.

It is understood that the above-mentioned method embodiments of the present disclosure can be combined with each other to form a combined embodiment without departing from the logic of the principle, which is limited by the space, and the detailed description of the present disclosure is omitted. Those skilled in the art will appreciate that in the above methods of the specific embodiments, the specific order of execution of the steps should be determined by their function and possibly their inherent logic.

In addition, the present disclosure also provides a computer vision model training device, an electronic device, a computer readable storage medium, and a program, which can be used to implement any one of the computer vision model training methods provided by the present disclosure, and the corresponding technical solutions and descriptions and corresponding descriptions in the methods section are not repeated.

Fig. 10 shows a schematic diagram of a computer vision model training apparatus according to an embodiment of the present disclosure, which includes a model determination module 100, an image generation module 101, and a model distillation module 102, as shown in fig. 10.

A model determination module 100 for obtaining a plurality of visual models;

an image generation module 101, configured to generate a plurality of synthetic images according to the visual models, respectively;

and a model distillation module 102, configured to perform model distillation on each of the synthesized images as an input of each of the visual models, so as to obtain a target visual model.

In one possible implementation, the image generation module 101 includes:

determining a first labeling result corresponding to each initial noise image;

In one possible implementation, the model distillation module 102 includes:

In one possible implementation, the model distillation submodule includes:

In one possible implementation, the apparatus further includes:

the model distillation submodule includes:

In some embodiments, functions of or modules included in the apparatus provided in the embodiments of the present disclosure may be used to execute the method described in the above method embodiments, and specific implementation thereof may refer to the description of the above method embodiments, and for brevity, will not be described again here.

Embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the above-mentioned method. The computer readable storage medium may be a volatile or non-volatile computer readable storage medium.

An embodiment of the present disclosure further provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the memory-stored instructions to perform the above-described method.

The disclosed embodiments also provide a computer program product comprising computer readable code or a non-transitory computer readable storage medium carrying computer readable code, which when run in a processor of an electronic device, the processor in the electronic device performs the above method.

The electronic device may be provided as a terminal, server, or other form of device.

Fig. 11 is a block diagram illustrating an electronic device 1100 in accordance with an example embodiment. For example, the electronic device 1100 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, or the like terminal.

Referring to fig. 11, electronic device 1100 may include one or more of the following components: processing component 1102, memory 1104, power component 1106, multimedia component 1108, audio component 1110, input/output (I/O) interface 1112, sensor component 1114, and communications component 1116.

The processing component 1102 generally controls the overall operation of the electronic device 1100, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing component 1102 may include one or more processors 1120 to execute instructions to perform all or a portion of the steps of the methods described above. Further, the processing component 1102 may include one or more modules that facilitate interaction between the processing component 1102 and other components. For example, the processing component 1102 may include a multimedia module to facilitate interaction between the multimedia component 1108 and the processing component 1102.

The memory 1104 is configured to store various types of data to support operations at the electronic device 1100. Examples of such data include instructions for any application or method operating on the electronic device 1100, contact data, phonebook data, messages, pictures, videos, and so forth. The memory 1104 may be implemented by any type or combination of volatile or non-volatile storage devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

The power supply component 1106 provides power to the various components of the electronic device 1100. The power components 1106 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 1100.

The multimedia component 1108 includes a screen that provides an output interface between the electronic device 1100 and a user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 1108 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the electronic device 1100 is in an operating mode, such as a shooting mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component 1110 is configured to output and/or input audio signals. For example, the audio component 1110 includes a Microphone (MIC) configured to receive external audio signals when the electronic device 1100 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signal may further be stored in the memory 1104 or transmitted via the communication component 1116. In some embodiments, the audio assembly 1110 further includes a speaker for outputting audio signals.

The I/O interface 1112 provides an interface between the processing component 1102 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor assembly 1114 includes one or more sensors for providing various aspects of state assessment for the electronic device 1100. For example, the sensor assembly 1114 may detect an open/closed state of the electronic device 1100, the relative positioning of components, such as a display and keypad of the electronic device 1100, the sensor assembly 1114 may also detect a change in the position of the electronic device 1100 or a component of the electronic device 1100, the presence or absence of user contact with the electronic device 1100, orientation or acceleration/deceleration of the electronic device 1100, and a change in the temperature of the electronic device 1100. The sensor assembly 1114 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 1114 may also include a light sensor, such as a Complementary Metal Oxide Semiconductor (CMOS) or Charge Coupled Device (CCD) image sensor, for use in imaging applications. In some embodiments, the sensor assembly 1114 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 1116 is configured to facilitate wired or wireless communication between the electronic device 1100 and other devices. The electronic device 1100 may access a wireless network based on a communication standard, such as a wireless network (WiFi), a second generation mobile communication technology (2G) or a third generation mobile communication technology (3G), or a combination thereof. In an exemplary embodiment, the communication component 1116 receives broadcast signals or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 1116 also includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the electronic device 1100 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 1104, is also provided that includes computer program instructions executable by the processor 1120 of the electronic device 1100 to perform the above-described method.

FIG. 12 is a block diagram illustrating another electronic device in accordance with an exemplary embodiment. For example, the electronic device 1200 may be provided as a server. Referring to fig. 12, electronic device 1200 includes a processing component 1222 that further includes one or more processors, and memory resources, represented by memory 1232, for storing instructions, such as applications, that are executable by processing component 1222. The application programs stored in memory 1232 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1222 is configured to execute instructions to perform the above-described methods.

The electronic device 1200 may also include a power supply component 1226 configured to perform power management of the electronic device 1200, a wired or wireless network interface 1250 configured to connect the electronic device 1200 to a network, and an input output (I/O) interface 1258. The electronic device 1200 may operate based on an operating system stored in the memory 1232, such as the Microsoft Server operating System (Windows Server)^TM) Apple Inc. of the present application based on the graphic user interface operating System (Mac OS X)^TM) Multi-user, multi-process computer operating system (Unix)^TM) Free and open native code Unix-like operating System (Linux)^TM) Open native code Unix-like operating System (FreeBSD)^TM) Or the like.

In an exemplary embodiment, a non-transitory computer readable storage medium, such as the memory 1232, is also provided that includes computer program instructions executable by the processing component 1222 of the electronic device 1200 to perform the above-described methods.

The present disclosure may be systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present disclosure.

The computer-readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The computer program product may be embodied in hardware, software or a combination thereof. In an alternative embodiment, the computer program product is embodied in a computer storage medium, and in another alternative embodiment, the computer program product is embodied in a Software product, such as a Software Development Kit (SDK), or the like.

The foregoing description of the various embodiments is intended to highlight various differences between the embodiments, and the same or similar parts may be referred to each other, and for brevity, will not be described again 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.

If the technical scheme of the application relates to personal information, a product applying the technical scheme of the application clearly informs personal information processing rules before processing the personal information, and obtains personal independent consent. If the technical scheme of the application relates to sensitive personal information, a product applying the technical scheme of the application obtains individual consent before processing the sensitive personal information, and simultaneously meets the requirement of 'express consent'. For example, at a personal information collection device such as a camera, a clear and significant identifier is set to inform that the personal information collection range is entered, the personal information is collected, and if the person voluntarily enters the collection range, the person is regarded as agreeing to collect the personal information; or on the device for processing the personal information, under the condition of informing the personal information processing rule by using obvious identification/information, obtaining personal authorization by modes of popping window information or asking a person to upload personal information of the person by himself, and the like; the personal information processing rule may include information such as a personal information processor, a personal information processing purpose, a processing method, and a type of personal information to be processed.

Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method of computer vision model training, the method comprising:

acquiring a plurality of visual models;

2. The method of claim 1, wherein generating a plurality of composite images from each of the visual models, respectively, comprises:

determining an initial plurality of noisy images;

3. The method of claim 2, wherein the first loss is used to characterize a loss generated by a corresponding visual model in detecting the noisy image;

4. The method of claim 3, wherein the determining of the first loss comprises:

determining an initial first labeling result corresponding to each noise image;

5. A method according to claim 3 or 4, characterized in that said second loss is determined from the values of the pixels in said noise image.

6. The method according to any one of claims 3 to 5, wherein the third loss is determined based on a first feature map output from each batch layer after the noise image is input into the corresponding visual model, and a second feature map input into each batch layer.

7. The method according to any one of claims 1-6, wherein performing model distillation using each of the composite images as an input to each of the visual models to obtain a target visual model comprises:

8. The method of claim 7, wherein performing model distillation using each of the composite images as input to the first visual model and each of the second visual models to obtain a target visual model comprises:

inputting each composite image into a first visual model to obtain a first detection result;

inputting each synthesized image into a second visual model respectively to obtain a second detection result;

9. The method according to any one of claims 1-8, further comprising:

10. The method of claim 8 or 9, wherein the second loss function is a sum of the L2 norms of the first detection result and each of the second detection results.

11. The method according to any one of claims 7-10, further comprising:

acquiring real images stored in the plurality of electronic devices;

the performing model distillation by using each of the synthetic images as the input of the first visual model and each of the second visual models to obtain a target visual model comprises:

12. A computer vision model training apparatus, the apparatus comprising:

a model determination module for obtaining a plurality of visual models;

an image generation module for generating a plurality of synthetic images according to the visual models respectively;

13. An electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to invoke the memory-stored instructions to perform the method of any of claims 1 to 11.

14. A computer readable storage medium having computer program instructions stored thereon, which when executed by a processor implement the method of any one of claims 1 to 11.