Self-training Optimization Method, Device, Electronic Device and Computer Readable Storage Medium

Abstract

The present application discloses a self-training optimization method, apparatus, electronic device, and computer-readable storage medium. The self-training optimization method includes: training an object detection inference model based on a training sample set determined according to acquired multimedia material data; evaluating the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model; screening sustainable training samples from the training sample set according to the model training value corresponding to each training sample, and adding them to newly acquired multimedia material data to update the training sample set of the object detection basic model; returning to execute the steps: training the object detection basic model according to the training sample set to obtain an object detection inference model, until the obtained object detection inference model meets the preset iterative training end condition, and taking the object detection inference model as the target inference model. The present application solves the technical problem of low model inference accuracy of the object detection inference model.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a self-training optimization method, device, electronic device, and computer-readable storage medium. Background Art

[0002] With the continuous development of artificial intelligence technology, the application of object detection models has become more and more extensive. For example, it can be applied to detect objects in images or audio. When training an object detection inference model, it is usually necessary to obtain a large amount of material data as sample data, and then train the object detection inference model according to a specific training method, such as incremental learning training with continuously incremented materials or batch learning training with all materials. However, the incremental learning training with continuously incremented materials or batch learning training with all materials will make the inference analysis effect of the trained object detection inference model gradually tend to be average or extreme, and neither the average nor the extreme of the inference analysis effect can better fit the user's on-site scenario, thus affecting the model inference accuracy of the object detection inference model. Summary of the Invention

[0003] The main purpose of this application is to provide a self-training optimization method, device, electronic device, and computer-readable storage medium, aiming to solve the technical problem of low model inference accuracy of the object detection inference model in the prior art.

[0004] To achieve the above purpose, this application provides a self-training optimization method, and the self-training optimization method includes:

[0005] Obtain multimedia material data, and determine a training sample set of an object detection basic model according to the multimedia material data;

[0006] Train the object detection basic model according to the training sample set to obtain an object detection inference model;

[0007] Evaluate the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model;

[0008] According to the model training value corresponding to each training sample, screen sustainable training samples in the training sample set and add them to newly obtained multimedia material data to update the training sample set of the object detection basic model;

[0009] Return to execute the steps: Train the object detection basic model according to the training sample set to obtain an object detection inference model, until the obtained object detection inference model meets the preset iteration training end condition, and use the object detection inference model as the target inference model.

[0010] Optionally, the step of evaluating the model training value of each training sample in the training sample set for the object detection inference model according to the preset sample evaluation model includes:

[0011] Obtain the sample acquisition time feature, the corresponding sample quality feature, and the corresponding user attention degree feature corresponding to the training sample;

[0012] Construct a sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature;

[0013] Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0014] Optionally, the step of constructing a sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature includes:

[0015] Obtain the class confidence feature corresponding to the class to which the training sample belongs, and the model inference deviation feature between the training sample in the object detection basic model and the object detection inference model;

[0016] Concatenate the sample acquisition time feature, the sample quality feature, the class confidence feature, the class confidence feature, and the model inference deviation feature to obtain the sample value evaluation feature.

[0017] Optionally, the training sample includes an image sample, and obtaining the class confidence feature corresponding to the class to which the training sample belongs includes:

[0018] Select each candidate region image suspected of having a target object in the image sample;

[0019] Classify each of the candidate region images to detect the probability of the target object existing in the candidate region image, and obtain each classification probability value;

[0020] Concatenate each of the classification probability values into the class confidence feature.

[0021] Optionally, obtaining the model inference deviation feature between the training sample in the object detection basic model and the object detection inference model includes:

[0022] Obtain the first inference result of the object detection basic model for the training sample and the second inference result of the object detection inference model for the training sample;

[0023] Construct the model inference deviation feature based on the inference deviation between the first inference result and the second inference result.

[0024] Optionally, the first inference result includes the first final inference result corresponding to the training sample and at least one corresponding first inference intermediate feature, and the second inference result includes the second final inference result corresponding to the training sample and at least one corresponding second inference intermediate feature. The step of constructing the model inference deviation feature based on the inference deviation between the first inference result and the second inference result includes:

[0025] Determine the intermediate feature inference deviation between each first inference intermediate feature and the corresponding second inference intermediate feature, and the final inference deviation between the first final inference result and the second final inference result;

[0026] Construct the model inference deviation feature based on each intermediate feature inference deviation and the final inference deviation.

[0027] Optionally, the step of evaluating the model training value of each training sample in the training sample set for the object detection inference model according to the preset sample evaluation model includes:

[0028] Obtain at least one dimensional feature corresponding to the training sample, where the dimensional feature at least includes one of a sample acquisition time feature, a sample quality feature, a class confidence feature corresponding to the class to which the training sample belongs, a model inference deviation feature between the training sample in the object detection base model and the object detection inference model, and a user attention degree feature;

[0029] Construct a sample value evaluation feature corresponding to the training sample according to the dimensional feature;

[0030] Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0031] This application also provides a self-training optimization device, which is applied to a self-training optimization device. The self-training optimization device includes:

[0032] A training sample set determination module, configured to obtain multimedia material data and determine a training sample set of an object detection base model according to the multimedia material data;

[0033] A model training module, configured to train the object detection base model according to the training sample set to obtain an object detection inference model;

[0034] A sample evaluation module, configured to evaluate the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model;

[0035] A training sample set update module, configured to screen sustainable training samples from the training sample set according to the model training value corresponding to each training sample, add them to newly acquired multimedia material data, so as to update the training sample set of the object detection basic model;

[0036] A continuous iteration module, configured to return and execute the steps: training an object detection basic model according to the training sample set to obtain an object detection inference model, until the obtained object detection inference model meets a preset iteration training end condition, and using the object detection inference model as the target inference model.

[0037] Optionally, the sample evaluation module is further configured to:

[0038] Obtain the sample acquisition time feature, the corresponding sample quality feature, and the corresponding user attention degree feature corresponding to the training sample;

[0039] Construct a sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature;

[0040] Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0041] Optionally, the sample evaluation module is further configured to:

[0042] Obtain the category confidence feature corresponding to the category to which the training sample belongs, and the model inference deviation feature between the training sample in the object detection basic model and the object detection inference model;

[0043] Concatenate the sample acquisition time feature, the sample quality feature, the category confidence feature, the category confidence feature, and the model inference deviation feature to obtain the sample value evaluation feature.

[0044] Optionally, the sample evaluation module is further configured to:

[0045] Select each candidate region image suspected of having a target object in the image sample;

[0046] Classify each of the candidate region images respectively to detect the probability of the target object existing in the candidate region image, and obtain each classification probability value;

[0047] Concatenate each of the classification probability values to form the class confidence feature.

[0048] Optionally, the sample evaluation module is further configured to:

[0049] Obtain a first inference result of the object detection basic model for the training sample and a second inference result of the object detection inference model for the training sample;

[0050] Construct the model inference deviation feature based on the inference deviation between the first inference result and the second inference result.

[0051] Optionally, the first inference result includes a first final inference result corresponding to the training sample and at least one first inference intermediate feature corresponding thereto, the second inference result includes a second final inference result corresponding to the training sample and at least one second inference intermediate feature corresponding thereto, and the sample evaluation module is further configured to:

[0052] Determine the intermediate feature inference deviation between each of the first inference intermediate features and the corresponding second inference intermediate feature and the final inference deviation between the first final inference result and the second final inference result;

[0053] Construct the model inference deviation feature based on each of the intermediate feature inference deviations and the final inference deviation.

[0054] Optionally, the sample evaluation module is further configured to:

[0055] Obtain at least one dimensional feature corresponding to the training sample, where the dimensional feature at least includes one of a sample acquisition time feature, a sample quality feature, a class confidence feature corresponding to the class to which the training sample belongs, a model inference deviation feature between the object detection basic model and the object detection inference model for the training sample, and a user attention degree feature;

[0056] Construct a sample value evaluation feature corresponding to the training sample according to the dimensional feature;

[0057] Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0058] The present application further provides an electronic device, which is a physical device. The electronic device includes: a memory, a processor, and a program of the self-training optimization method stored on the memory and executable on the processor. When the program of the self-training optimization method is executed by the processor, the steps of the self-training optimization method as described above can be implemented.

[0059] The present application also provides a computer-readable storage medium, on which a program for implementing the self-training optimization method is stored. When the program of the self-training optimization method is executed by a processor, the steps of the self-training optimization method as described above are implemented.

[0060] The present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the self-training optimization method as described above are implemented.

[0061] The present application provides a self-training optimization method, device, electronic device, and computer-readable storage medium. First, multimedia material data is obtained, and a training sample set is determined according to the multimedia material data; an object detection basic model is trained according to the training sample set to obtain an object detection inference model; according to a preset sample evaluation model, the model training value of each training sample in the training sample set for the object detection inference model is evaluated; according to the model training value corresponding to each training sample, sustainable training samples are screened out in the training sample set; new multimedia material data is obtained, and the training sample set is updated according to the new multimedia material data and the sustainable training samples. Therefore, after each training in the present application, sustainable training samples will be selected from the training sample set of the current training according to the model training value of the training samples and added to the new multimedia material data collected in the next stage as the training sample set for the next training, and then the steps are returned to be executed: an object detection basic model is trained according to the training sample set to obtain an object detection inference model for the next training. Therefore, the present application can realize continuous iterative training of the object detection inference model according to the multimedia material data collected in different stages, and since there are differences in the user site scenarios where the training materials are collected each time, the object detection inference model will continuously adapt to different user site scenarios during continuous iterative training, so that the object detection inference model that finally meets the preset model test conditions will not be unable to fit the user site scenario due to the inference analysis effect gradually tending to be average or extreme, thereby improving the model inference accuracy of the object detection inference model. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0064] Figure 1Schematic flowchart of the first embodiment of the self-training optimization method of the present application;

[0065] Figure 2 Schematic flowchart of the process of training an object detection inference model in the self-training optimization method of the present application;

[0066] Figure 3 Schematic flowchart of the second embodiment of the self-training optimization method of the present application;

[0067] Figure 4 Schematic flowchart of the third embodiment of the self-training optimization method of the present application;

[0068] Figure 5 Schematic flowchart of the fourth embodiment of the self-training optimization method of the present application;

[0069] Figure 6 Schematic diagram of the device structure of the hardware operating environment involved in the self-training optimization method in the embodiments of the present application.

[0070] The implementation, functional features, and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. Detailed implementation manners

[0071] To make the above objects, features, and advantages of the present application more obvious and understandable, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0072] The embodiments of the present application provide a self-training optimization method. In an embodiment of the self-training optimization method of the present application, refer to Figure 1 , the self-training optimization method includes:

[0073] Step S10: Obtain multimedia material data, and determine a training sample set for an object detection basic model according to the multimedia material data;

[0074] Step S20: Train the object detection basic model according to the training sample set to obtain an object detection inference model;

[0075] Step S30: Evaluate the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model;

[0076] Step S40: According to the model training value corresponding to each training sample, screen sustainable training samples from the training sample set and add them to the newly obtained multimedia material data to update the training sample set of the object detection basic model.

[0077] Step S50: Return to execute the step of training the object detection basic model according to the training sample set to obtain an object detection inference model until the obtained object detection inference model meets the preset iterative training end condition, and use the object detection inference model as the target inference model.

[0078] In this embodiment, it should be noted that the multimedia material data may be multimedia data such as image data or sound data. The object detection basic model may be a teacher model for object detection, and the object detection inference model may be a student model for object detection. Object detection may specifically be object behavior detection, object recognition detection, or object classification detection, etc. The model training value is the value of the training sample for training the object detection inference model. The higher the model training value, the more beneficial the training sample is to the training of the object detection inference model, and it can be used to improve the adaptation degree of the obtained object detection inference model to different user on-site scenarios.

[0079] As an example, steps S10 to S50 include: obtaining multimedia material data, and determining a training sample set of an object detection basic model according to the multimedia material data; training the object detection basic model according to the training sample set to obtain an object detection inference model; obtaining dimension features of each training sample in the training sample set under different evaluation dimensions, and evaluating the model training value of each training sample for the object detection inference model by respectively inputting the dimension features corresponding to each training sample into a preset sample evaluation model; screening, according to the model training value corresponding to each training sample, training samples in the training sample set whose model training value is greater than a preset training value threshold as sustainable training samples; obtaining new multimedia material data, and using the sustainable training samples and the new multimedia material data together as a training sample set for the next training; returning to execute the steps: training the object detection basic model according to the training sample set to obtain an object detection inference model, and performing the next training with the object detection basic model; if it is detected that the obtained object detection inference model meets a preset iteration training end condition, using the object detection inference model as a target inference model. Among them, since there are differences in the user's real scenarios when obtaining multimedia material data each time, the purpose of training the object detection inference model with multimedia material data collected in different stages is realized during the continuous iterative training of the object detection inference model, so that the finally trained object detection inference model can adapt to different user real scenarios. Among them, the evaluation dimension may be one or more of a time dimension, a sample quality dimension, a category confidence dimension corresponding to the category to which the sample belongs, a model inference deviation dimension of the object detection basic model and the object detection inference model for inferring the sample, and a user attention degree dimension.

[0080] As an example, the determining a training sample set of an object detection basic model according to the multimedia material data includes:

[0081] If this training is the first training of the object detection inference model, directly using the multimedia material data as the training sample set; if this training is not the first training of the object detection inference model, obtaining the sustainable training samples screened during the previous training, and using the sustainable training samples screened during the previous training and the multimedia material data together as the training sample set.

[0082] As an example, the step of training the object detection basic model according to the training sample set to obtain an object detection inference model includes:

[0083] Obtain training samples from the training sample set. By inputting the training samples into the object detection base model, perform object detection on the training samples to obtain the first object detection result, and obtain the first inference intermediate feature of the object detection base model for the training samples, where the first inference intermediate feature is the output of the hidden layer of the object detection base model after the training samples are input; by inputting the training samples into the object detection model to be trained, perform object detection on the training samples to obtain the second object detection result, and obtain the second inference intermediate feature of the object detection model to be trained for the training samples, where the second inference intermediate feature is the output of the hidden layer of the object detection model to be trained after the training samples are input; construct the model loss corresponding to the object detection model to be trained based on the difference between the first object detection result and the second object detection result, and the difference between the first inference intermediate feature and the second inference intermediate feature; determine whether the model loss converges. If it converges, use the object detection model to be trained as the object detection inference model. If it does not converge, update the object detection model to be trained according to the gradient calculated by the model loss, and return to execute the steps: obtain training samples from the training sample set to perform the next round of training on the object detection model to be trained until the calculated model loss converges. Wherein, the object detection model to be trained is an object detection inference model that has not been trained well.

[0084] Wherein, the step of evaluating the model training value of each training sample in the training sample set for the object detection inference model according to the preset sample evaluation model includes:

[0085] Step A10, obtain at least one dimensional feature corresponding to the training sample, where the dimensional feature at least includes one of a sample acquisition time feature, a sample quality feature, a class confidence feature corresponding to the class to which the training sample belongs, a model inference deviation feature between the training sample in the object detection base model and the object detection inference model, and a user attention degree feature;

[0086] Step A20, construct a sample value evaluation feature corresponding to the training sample according to the dimensional feature;

[0087] Step A30, evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0088] In this embodiment, it should be noted that the sample acquisition time feature is a feature representing the distance between the sample acquisition time and the current time; the sample quality feature is a feature representing the quality of the sample. For example, if the training sample is an image, the sample quality feature can be a feature vector composed of feature values such as image clarity, image resolution, and the degree of target occlusion in the image; the class confidence feature is a feature representing the confidence of each object in the training sample being the target object. For example, if the training sample is an image and there are 10 objects in the image suspected to be human objects, the feature vector composed of the probability values of the 10 objects being human objects is the class confidence feature; the model inference deviation feature is a feature representing the deviation between the object detection base model and the object detection inference model during the model inference process; the user attention degree feature is a feature representing the level of user attention. For example, the value of the user attention degree feature can be set from 0 to 1, where 0 indicates that the user does not pay attention to the training sample, and 1 indicates that the user's attention to the training sample is 100%.

[0089] As an example, steps A10 to A30 include: obtaining at least one dimensional feature corresponding to the training sample, where the dimensional feature at least includes one of the sample acquisition time feature, the sample quality feature, the class confidence feature corresponding to the class to which the training sample belongs, the model inference deviation feature between the training sample in the object detection base model and the object detection inference model, and the user attention degree feature; splicing the dimensional features according to a preset splicing order to obtain the sample value evaluation feature corresponding to the training sample; inputting the sample value evaluation feature into the preset sample evaluation model for scoring, and using the scoring value output by the sample value evaluation model as the model training value of the training sample for the object detection inference model. The embodiment of the present application achieves the purpose of evaluating the model training value of the training sample from multiple dimensions such as the time dimension, the sample quality dimension, the class confidence dimension corresponding to the class to which the sample belongs, the model inference deviation dimension of the sample inference by the object detection base model and the object detection inference model, and the user attention degree dimension, provides sufficient decision-making basis for the evaluation of the model training value, and therefore improves the accuracy of the evaluation of the model training value of the training sample.

[0090] As an example, referring to Figure 2 , Figure 2This is a schematic flowchart for training an object detection inference model in an embodiment of the present application. Among them, the basic algorithm model is the object detection basic model, the algorithm models output in the first to the Nth training are all the object detection inference models, the evaluation model is the preset sample evaluation model, the training material is the multimedia material data, and the evaluation model is used to screen out sustainable learning training materials in the training material during each training process, that is, to screen out sustainable training samples, and add the sustainable training samples to the training material collected in the next training as the training sample set of the object detection inference model. Thus, during the entire training process of the object detection inference model, the training material can be continuously selected, and sustainable learning training materials that can be used for continuous learning training can be selected, so that the value of the training material for each training is higher and the scene coverage is wider. Therefore, the effect of the trained object detection inference model is more stable and the adaptability is stronger, and there will be no situation where the trained object detection inference model does not match the user's real scene and affects the inference accuracy of the object detection inference model. Therefore, the model inference accuracy of the object detection inference model is improved.

[0091] The embodiment of the present application provides a self-training optimization method. First, multimedia material data is obtained, and a training sample set is determined according to the multimedia material data; an object detection basic model is trained according to the training sample set to obtain an object detection inference model; according to a preset sample evaluation model, the model training value of each training sample in the training sample set for the object detection inference model is evaluated; according to the model training value corresponding to each training sample, sustainable training samples are screened out in the training sample set; new multimedia material data is obtained, and the training sample set is updated according to the new multimedia material data and the sustainable training samples. Therefore, after each training in the embodiment of the present application, sustainable training samples will be selected from the training sample set of the current training according to the model training value of the training samples and added to the new multimedia material data collected in the next stage as the training sample set for the next training, and then the steps are returned to execute: training the object detection basic model according to the training sample set to obtain an object detection inference model for the next training. Therefore, the embodiment of the present application can realize continuous iterative training of the object detection inference model according to the multimedia material data collected in different stages. And because there will be differences in the user site scenarios for collecting training materials each time, the object detection inference model will continuously adapt to different user site scenarios during continuous iterative training, so that the object detection inference model that finally meets the preset model test conditions will not fail to fit the user site scenario because the inference analysis effect gradually tends to be average or extreme, thereby improving the model inference accuracy of the object detection inference model.

[0092] Further, referring to Figure 3, in another embodiment of the present application, for the same or similar content as in the above embodiment, reference may be made to the above introduction and will not be elaborated hereinafter. On this basis, the step of evaluating the model training value of each training sample in the training sample set for the object detection inference model according to the preset sample evaluation model includes:

[0093] Step S31, obtaining the sample acquisition time feature, the corresponding sample quality feature, and the corresponding user attention degree feature corresponding to the training sample;

[0094] Step S32, constructing a sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature;

[0095] Step S33, evaluating the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0096] As an example, steps S31 to S33 include: obtaining the sample acquisition time corresponding to the training sample, calculating the time interval length between the sample acquisition time and the current time, and using the time interval length as the sample acquisition time feature; obtaining the sample quality feature corresponding to the training sample; obtaining the user attention degree label corresponding to the training sample, and using the user attention degree label as the user attention degree feature, where the user attention degree label can be an identifier used by the user to identify the attention degree of the user; splicing the sample acquisition time feature, the sample quality feature, and the user attention degree label to obtain a sample value evaluation feature, inputting the sample value evaluation feature into the preset sample evaluation model for scoring, and using the scoring value output by the sample evaluation model as the model training value of the training sample for the object detection inference model.

[0097] As an example, the training sample can be an image sample, and the obtaining the sample quality feature corresponding to the training sample includes:

[0098] obtaining the image clarity, the corresponding image resolution, and the occlusion degree of the target object in the image sample, where the target object can be a human object or a specific item object; splicing the image clarity, the image resolution, and the occlusion degree to obtain a sample quality feature.

[0099] As an example, the training sample can be a sound sample, and the obtaining the sample quality feature corresponding to the training sample includes:

[0100] Obtain the sound amplitude, corresponding sound frequency, and corresponding noise interference degree corresponding to the sound sample; splice the sound amplitude, sound frequency, and the noise interference degree to obtain the sample quality feature.

[0101] The embodiment of the present application provides a method for evaluating the value of model training, that is, obtaining the sample acquisition time feature, corresponding sample quality feature, and corresponding user attention degree feature corresponding to the training sample; constructing the sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature, so as to realize constructing the sample value evaluation feature according to the proximity of the sample acquisition time, the sample quality, and the user's attention degree to the training sample, making the sample value evaluation feature have sufficient feature information, and then inputting the sample value evaluation feature into the preset sample evaluation model to evaluate the model training value of the training sample for the object detection inference model, which can provide a sufficient decision-making basis for the evaluation of the model training value of the training sample and ensure the accuracy of the evaluation of the model training value of the training sample.

[0102] Further, referring to Figure 4 , in another embodiment of the present application, the same or similar content as the above embodiment can be referred to the above introduction and will not be repeated hereinafter. On this basis, the step of constructing the sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature includes:

[0103] Step S321, obtain the class confidence feature corresponding to the class to which the training sample belongs, and the model inference deviation feature between the training sample in the object detection basic model and the object detection inference model;

[0104] Step S322, splice the sample acquisition time feature, the sample quality feature, the class confidence feature, the class confidence feature, and the model inference deviation feature to obtain the sample value evaluation feature.

[0105] As an example, steps S321 to S322 include: classifying each candidate object in the training sample to predict the probability that the candidate object is the target object, obtaining the classification probability corresponding to each target object; splicing the classification probabilities to obtain a class confidence feature; obtaining a first inference result of the object detection basic model for the training sample and a second inference result of the object detection inference model for the training sample, calculating an inference result deviation value between the first inference result and the second inference result, and using the inference result deviation value as a model inference deviation feature; splicing the sample acquisition time feature, the sample quality feature, the class confidence feature, the class confidence feature, and the model inference deviation feature to obtain the sample value evaluation feature.

[0106] Among them, the training sample includes an image sample, and obtaining the class confidence feature corresponding to the class to which the training sample belongs includes:

[0107] Step B10, selecting, in the image sample, each candidate region image where a target object is suspected to exist;

[0108] Step B20, respectively classifying each candidate region image to detect the probability that the target object exists in the candidate region image, obtaining each classification probability value;

[0109] Step B30, splicing each classification probability value into the class confidence feature.

[0110] As an example, steps B10 to B30 include: performing object detection on the image sample, and framing each image region where a target object is suspected to exist in the image sample as a candidate region image; respectively classifying each candidate region image to predict the probability that the target object exists in each candidate region image, obtaining each classification probability value corresponding to each candidate region image; using the feature vector obtained by splicing each classification probability value as the class confidence feature.

[0111] The embodiment of the present application provides a method for constructing sample value evaluation features, that is, obtaining the class confidence feature corresponding to the class to which the training sample belongs, and the model inference deviation feature between the object detection basic model and the object detection inference model of the training sample; splicing the sample acquisition time feature, the sample quality feature, the class confidence feature, the class confidence feature and the model inference deviation feature to obtain the sample value evaluation feature, achieving the purpose of constructing the sample value evaluation feature based on the proximity of the sample acquisition time, the sample quality, the attention degree of the user to the corresponding training sample, the class confidence corresponding to the required class of the training sample, and the model inference deviation between the object detection basic model and the object detection inference model of the training sample, so that the sample value evaluation feature has more sufficient feature information. Furthermore, the accuracy of the model value evaluation of the training sample based on the sample value evaluation feature with more sufficient feature information will be higher. Therefore, it lays a foundation for improving the accuracy of the model value evaluation of the training sample.

[0112] Further, referring to Figure 5 , in another embodiment of the present application, the same or similar content as the above embodiment can be referred to the above introduction and will not be elaborated hereinafter. On this basis, obtaining the model inference deviation feature between the object detection basic model and the object detection inference model of the training sample includes:

[0113] Step C10, obtaining the first inference result of the object detection basic model for the training sample and the second inference result of the object detection inference model for the training sample;

[0114] Step C20, constructing the model inference deviation feature according to the inference deviation between the first inference result and the second inference result.

[0115] In this embodiment, it should be noted that the first inference result may be one or more of the final inference result or the intermediate inference result output by the object basic model for the training sample; the second inference result may be one or more of the final inference result or the intermediate inference result output by the object inference model for the training sample. Among them, the final inference result may be the output value of the output layer of the model, and the intermediate inference result may be the intermediate feature output by the intermediate hidden layer of the model.

[0116] As an example, steps C10 to C20 include: obtaining the output value of the output layer of the object detection basic model after the training sample is input into the object detection basic model, and using the output value of the output layer of the object detection basic model as the first inference result; obtaining the output value of the output layer of the object detection basic model after the training sample is input into the object detection inference model, and using the output value of the output layer of the object detection inference model as the second inference result; calculating the difference between the first inference result and the second inference result to obtain the final inference deviation, and using the final inference deviation as the model inference deviation feature.

[0117] In another implementation, steps C10 to C20 include: obtaining the intermediate features output by the hidden layer of the object detection basic model after the training sample is input into the object detection basic model, and using the intermediate features output by the hidden layer of the object detection basic model as the first inference result; obtaining the intermediate features output by the hidden layer of the object detection basic model after the training sample is input into the object detection inference model, and using the intermediate features output by the hidden layer of the object detection inference model as the second inference result; calculating the difference between the first inference result and the second inference result to obtain the intermediate feature inference deviation, and using the intermediate feature inference deviation as the model inference deviation feature.

[0118] Wherein, the first inference result includes the first final inference result corresponding to the training sample and at least one corresponding first inference intermediate feature, the second inference result includes the second final inference result corresponding to the training sample and at least one corresponding second inference intermediate feature, and constructing the model inference deviation feature based on the inference deviation between the first inference result and the second inference result includes:

[0119] Step C21, determining the intermediate feature inference deviation between each first inference intermediate feature and the corresponding second inference intermediate feature and the final inference deviation between the first final inference result and the second final inference result;

[0120] Step C22, constructing the model inference deviation feature based on each intermediate feature inference deviation and the final inference deviation.

[0121] In this embodiment, it should be noted that the number of the first inference intermediate features and the second inference intermediate features can be one or more.

[0122] As an example, steps C21 to C22 include: obtaining first inference intermediate features output by each preset hidden layer of the object detection base model for the training sample, and a first final inference result output by the output layer of the object detection base model for the training sample; obtaining second inference intermediate features output by each preset hidden layer of the object detection inference model for the training sample, and a second final inference result output by the output layer of the object detection inference model for the training sample; calculating the difference between each first inference intermediate feature and the corresponding second inference intermediate feature to obtain intermediate feature differences, aggregating the intermediate feature differences to obtain an intermediate feature inference deviation; calculating the difference between the first final inference result and the second final inference result to obtain a final inference deviation; splicing the intermediate feature inference deviation and the final inference deviation to obtain a model inference deviation feature. For example, assuming that the intermediate feature inference deviation is a matrix and the final inference deviation is a numerical value, a vector corresponding to the final inference deviation can be constructed in advance by inserting 0s, and then the vector is spliced to the matrix as the intermediate feature inference deviation to obtain a new matrix as the model inference deviation feature. Alternatively, the intermediate feature inference deviation can be mapped from a matrix to a vector of a preset length in advance, and then the numerical value as the final inference deviation is spliced to the vector of the preset length to obtain a new vector as the model inference deviation feature. The preset hidden layer is a hidden layer with a preset network layer depth. For example, the 100th layer network, 500th layer network, and 1000th layer network of the object detection base model can be selected as each preset hidden layer.

[0123] As an example, the intermediate feature difference is a matrix, and the aggregating of the intermediate feature differences to obtain an intermediate feature inference deviation includes:

[0124] Mapping each intermediate feature difference to a preset dimensional space to obtain standard intermediate feature difference matrices; adding up the standard intermediate feature difference matrices to obtain an intermediate feature inference deviation.

[0125] In another implementable manner, the intermediate feature difference is a matrix, and the aggregating of the intermediate feature differences to obtain an intermediate feature inference deviation includes:

[0126] Mapping each intermediate feature difference to a preset dimensional space to obtain standard intermediate feature difference matrices; weighted averaging the standard intermediate feature difference matrices to obtain an intermediate feature inference deviation.

[0127] An embodiment of the present application provides a method for constructing a model inference feature deviation. First, when performing model inference, each first inference intermediate feature and the first final inference result generated during the inference training of the object detection basic model, as well as each second inference intermediate feature and the second final inference result generated during the inference training of the object detection inference model are obtained, so as to determine the intermediate feature inference deviation between each of the first inference intermediate features and the corresponding second inference intermediate features and the final inference deviation between the first final inference result and the second final inference result; based on each of the intermediate feature inference deviations and the final inference deviation, the model inference deviation feature is constructed, thus achieving the purpose of constructing the model inference deviation feature based on the differences between the object detection basic model and the object detection inference model during the entire inference process, rather than only evaluating the model inference deviation based on the final output values of the object detection basic model and the object detection inference model. Therefore, sufficient information is provided for the construction of the model inference deviation feature, enabling the model inference deviation feature to more fully and accurately represent the model inference deviation between the object detection basic model and the object detection inference model. The model inference deviation feature has richer feature information, so that the accuracy of evaluating the model training value of the training samples based on the model inference deviation feature is higher, laying a foundation for improving the accuracy of evaluating the model training value of the training samples.

[0128] An embodiment of the present application further provides a self-training optimization device. The self-training optimization device is applied to a self-training optimization device, and the self-training optimization device includes:

[0129] A training sample set determination module, configured to obtain multimedia material data and determine a training sample set of an object detection basic model according to the multimedia material data;

[0130] A model training module, configured to train the object detection basic model according to the training sample set to obtain an object detection inference model;

[0131] A sample evaluation module, configured to evaluate the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model;

[0132] A training sample set update module, configured to screen sustainable training samples from the training sample set according to the model training value corresponding to each training sample and add them to newly obtained multimedia material data to update the training sample set of the object detection basic model;

[0133] A continuous iteration module, configured to return and execute the steps: training the object detection basic model according to the training sample set to obtain an object detection inference model until the obtained object detection inference model meets a preset iteration training end condition, and using the object detection inference model as the target inference model.

[0134] Optionally, the sample evaluation module is further configured to:

[0135] Obtain the sample collection time feature, the corresponding sample quality feature, and the corresponding user attention degree feature corresponding to the training sample;

[0136] Construct a sample value evaluation feature corresponding to the training sample based on the sample collection time feature, the sample quality feature, and the user attention degree feature;

[0137] Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0138] Optionally, the sample evaluation module is further configured to:

[0139] Obtain the class confidence feature corresponding to the class to which the training sample belongs, and the model inference deviation feature between the object detection basic model and the object detection inference model;

[0140] Concatenate the sample collection time feature, the sample quality feature, the class confidence feature, the class confidence feature, and the model inference deviation feature to obtain the sample value evaluation feature.

[0141] Optionally, the sample evaluation module is further configured to:

[0142] Select candidate region images suspected of having a target object in the image sample;

[0143] Classify each of the candidate region images to detect the probability of the target object existing in the candidate region image, and obtain respective classification probability values;

[0144] Concatenate the respective classification probability values into the class confidence feature.

[0145] Optionally, the sample evaluation module is further configured to:

[0146] Obtain a first inference result of the object detection basic model for the training sample and a second inference result of the object detection inference model for the training sample;

[0147] Construct the model inference deviation feature based on the inference deviation between the first inference result and the second inference result.

[0148] Optionally, the first inference result includes a first final inference result corresponding to the training sample and at least one corresponding first intermediate inference feature, and the second inference result includes a second final inference result corresponding to the training sample and at least one corresponding second intermediate inference feature. The sample evaluation module is further configured to:

[0149] Determine the intermediate feature inference deviation between each of the first intermediate inference features and the corresponding second intermediate inference feature, and the final inference deviation between the first final inference result and the second final inference result;

[0150] Construct the model inference deviation feature based on each of the intermediate feature inference deviations and the final inference deviation.

[0151] Optionally, the sample evaluation module is further configured to:

[0152] Obtain at least one dimensional feature corresponding to the training sample, where the dimensional feature at least includes one of a sample acquisition time feature, a sample quality feature, a class confidence feature corresponding to the class to which the training sample belongs, a model inference deviation feature between the object detection base model and the object detection inference model for the training sample, and a user attention degree feature;

[0153] Construct a sample value evaluation feature corresponding to the training sample according to the dimensional feature;

[0154] Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

[0155] The self-training optimization device provided in this application adopts the self-training optimization method in the above embodiment, and solves the technical problem of low model inference accuracy of the object detection inference model. Compared with the prior art, the beneficial effects of the self-training optimization device provided in the embodiments of this application are the same as those of the self-training optimization method provided in the above embodiment, and other technical features in the self-training optimization device are the same as the features disclosed in the method of the above embodiment, and will not be elaborated here.

[0156] An embodiment of this application provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the self-training optimization method in the first embodiment above.

[0157] Next, refer to Figure 6, which shows a schematic structural diagram of an electronic device suitable for implementing the embodiments of the present disclosure. The electronic devices in the embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The electronic device shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0158] As Figure 6 shown, the electronic device may include a processing device (such as a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). In the RAM, various programs and data required for the operation of the electronic device are also stored. The processing device, the ROM, and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0159] Generally, the following systems may be connected to the I / O interface: input devices including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices including, for example, magnetic tapes, hard disks, etc.; and a communication device. The communication device may allow the electronic device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows an electronic device having various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or had alternatively.

[0160] Specifically, according to the embodiments of the present disclosure, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, the embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from the network through the communication device, or installed from the storage device, or installed from the ROM. When the computer program is executed by the processing device, the above functions defined in the methods of the embodiments of the present disclosure are performed.

[0161] The electronic device provided by this application adopts the self-training optimization method in the above-mentioned embodiment, and solves the technical problem of low model inference accuracy of the object detection inference model. Compared with the prior art, the beneficial effects of the electronic device provided by the embodiment of this application are the same as those of the self-training optimization method provided by the above-mentioned embodiment, and other technical features in this electronic device are the same as those disclosed in the method of the above-mentioned embodiment, which will not be elaborated here.

[0162] It should be understood that each part of the present disclosure can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0163] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0164] This embodiment provides a computer-readable storage medium having computer-readable program instructions stored thereon, and the computer-readable program instructions are used to execute the method of the self-training optimization method in the first embodiment above.

[0165] The computer-readable storage medium provided by the embodiment of this application can be, for example, a USB flash drive, but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or combined with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0166] The above computer-readable storage medium can be included in the electronic device; it can also exist separately without being assembled into the electronic device.

[0167] The above computer-readable storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to: obtain multimedia material data, and determine a training sample set for an object detection basic model according to the multimedia material data; train the object detection basic model according to the training sample set to obtain an object detection inference model; evaluate the model training value of each training sample in the training sample set according to a preset sample evaluation model; according to the model training value corresponding to each training sample, screen sustainable training samples in the training sample set and add them to newly obtained multimedia material data to update the training sample set of the object detection basic model; return to execute the steps: train the object detection basic model according to the training sample set to obtain an object detection inference model, until the obtained object detection inference model meets the preset iteration training end condition, and use the object detection inference model as the target inference model.

[0168] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a 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 may be connected to an external computer (e.g., by connecting through the Internet service provider via the Internet).

[0169] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0170] The modules involved in the embodiments of the present disclosure can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation on the unit itself in some cases.

[0171] The computer-readable storage medium provided by this application stores computer-readable program instructions for executing the above self-training optimization method, and solves the technical problem of low model inference accuracy of the object detection inference model. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the embodiments of this application are the same as those of the self-training optimization method provided by the above embodiments, and will not be elaborated here.

[0172] This application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the self-training optimization method as described above.

[0173] The computer program product provided by this application solves the technical problem of low model inference accuracy of the object detection inference model. Compared with the prior art, the beneficial effects of the computer program product provided by the embodiments of this application are the same as those of the self-training optimization method provided by the above embodiments, and will not be elaborated here.

[0174] The above are only the preferred embodiments of this application, and do not limit the patent scope of this application accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of this application, or directly or indirectly applied in other related technical fields, shall be included in the patent scope of this application by the same token.

Claims

1. A self-training optimization method, characterized in that, The self-training optimization method includes: Obtain multimedia material data, and determine a training sample set of the object detection basic model according to the multimedia material data; Train the object detection basic model according to the training sample set to obtain an object detection inference model; Evaluate the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model; According to the model training value corresponding to each training sample, screen sustainable training samples in the training sample set and add them to newly obtained multimedia material data to update the training sample set of the object detection basic model; Return to execute the steps: Train the object detection basic model according to the training sample set to obtain an object detection inference model, until the obtained object detection inference model meets the preset iteration training end condition, and use the object detection inference model as the target inference model; Among them, the step of evaluating the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model includes: Obtain the sample collection time feature, the corresponding sample quality feature, and the corresponding user attention degree feature corresponding to the training sample; Construct a sample value evaluation feature corresponding to the training sample according to the sample collection time feature, the sample quality feature, and the user attention degree feature; Evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

2. The self-training optimization method according to claim 1, wherein The step of constructing a sample value evaluation feature corresponding to the training sample according to the sample collection time feature, the sample quality feature, and the user attention degree feature includes: Obtain the class confidence feature corresponding to the class to which the training sample belongs, and the model inference deviation feature between the training sample in the object detection basic model and the object detection inference model; Concatenate the sample collection time feature, the sample quality feature, the class confidence feature, the class confidence feature, and the model inference deviation feature to obtain the sample value evaluation feature.

3. The self-training optimization method according to claim 2, wherein The training sample includes an image sample, and obtaining the class confidence feature corresponding to the class to which the training sample belongs includes: Select each candidate region image suspected of having a target object in the image sample; Classify each candidate region image respectively to detect the probability of the target object existing in the candidate region image to obtain each classification probability value; Concatenate each classification probability value into the class confidence feature.

4. The self-training optimization method according to claim 2, wherein Obtaining the model inference deviation feature between the training sample in the object detection basic model and the object detection inference model includes: Obtain the first inference result of the object detection basic model for the training sample and the second inference result of the object detection inference model for the training sample; Construct the model inference deviation feature according to the inference deviation between the first inference result and the second inference result.

5. The self-training optimization method according to claim 4, wherein, The first inference result includes the first final inference result corresponding to the training sample and at least one corresponding first intermediate inference feature, and the second inference result includes the second final inference result corresponding to the training sample and at least one corresponding second intermediate inference feature. Constructing the model inference deviation feature based on the inference deviation between the first inference result and the second inference result includes: Determining the intermediate feature inference deviation between each first intermediate inference feature and the corresponding second intermediate inference feature, and the final inference deviation between the first final inference result and the second final inference result; Constructing the model inference deviation feature based on each intermediate feature inference deviation and the final inference deviation.

6. The self-training optimization method according to claim 1, wherein The step of evaluating the model training value of each training sample in the training sample set for the object detection inference model according to the preset sample evaluation model includes: Obtaining at least one dimensional feature corresponding to the training sample, where the dimensional feature at least includes one of a sample acquisition time feature, a sample quality feature, a class confidence feature corresponding to the class to which the training sample belongs, a model inference deviation feature between the training sample in the object detection basic model and the object detection inference model, and a user attention degree feature; Constructing a sample value evaluation feature corresponding to the training sample according to the dimensional feature; Evaluating the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

7. A self-training optimization device, characterized in that The self-training optimization device includes: A training sample set determination module, configured to obtain multimedia material data and determine a training sample set of an object detection basic model according to the multimedia material data; A model training module, configured to train the object detection basic model according to the training sample set to obtain an object detection inference model; A sample evaluation module, configured to evaluate the model training value of each training sample in the training sample set for the object detection inference model according to a preset sample evaluation model; A training sample set update module, configured to screen sustainable training samples from the training sample set according to the model training value corresponding to each training sample and add them to newly obtained multimedia material data to update the training sample set of the object detection basic model; A continuous iteration module, configured to return and execute the steps: training the object detection basic model according to the training sample set to obtain an object detection inference model until the obtained object detection inference model meets a preset iteration training end condition, and using the object detection inference model as a target inference model; The sample evaluation module is further configured to obtain the sample acquisition time feature, the corresponding sample quality feature, and the corresponding user attention degree feature corresponding to the training sample; construct a sample value evaluation feature corresponding to the training sample according to the sample acquisition time feature, the sample quality feature, and the user attention degree feature; and evaluate the model training value of the training sample for the object detection inference model by inputting the sample value evaluation feature into the preset sample evaluation model.

8. An electronic device, characterized in that, The electronic device includes: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the steps of the self-training optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, A program for implementing the self-training optimization method is stored on the computer-readable storage medium, and when the program for implementing the self-training optimization method is executed by a processor, the steps of the self-training optimization method according to any one of claims 1 to 6 are implemented.

Images