WO2020155609A1 - 一种目标对象处理方法、装置、电子设备及存储介质 - Google Patents

一种目标对象处理方法、装置、电子设备及存储介质 Download PDF

Info

Publication number
WO2020155609A1
WO2020155609A1 PCT/CN2019/101448 CN2019101448W WO2020155609A1 WO 2020155609 A1 WO2020155609 A1 WO 2020155609A1 CN 2019101448 W CN2019101448 W CN 2019101448W WO 2020155609 A1 WO2020155609 A1 WO 2020155609A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
processing module
scene
processing
neural network
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2019/101448
Other languages
English (en)
French (fr)
Chinese (zh)
Inventor
韩世欣
郭宇
秦红伟
赵钰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Sensetime Technology Development Co Ltd
Original Assignee
Beijing Sensetime Technology Development Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Sensetime Technology Development Co Ltd filed Critical Beijing Sensetime Technology Development Co Ltd
Priority to KR1020207018377A priority Critical patent/KR20200096556A/ko
Priority to JP2020533136A priority patent/JP7064593B2/ja
Priority to SG11202005886RA priority patent/SG11202005886RA/en
Priority to US16/901,190 priority patent/US11403489B2/en
Publication of WO2020155609A1 publication Critical patent/WO2020155609A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/776—Validation; Performance evaluation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection

Definitions

  • the present disclosure relates to the field of computer vision technology, and in particular to a target object processing method, device, electronic equipment and storage medium.
  • training data is obtained through model training in the cloud.
  • the available training data is different from the data of the actual application scenario, and the detection requirements of different application scenarios are different. Therefore, the model training is performed in the cloud.
  • the training data obtained is not suitable for all application scenarios.
  • the present disclosure proposes a technical solution for target object processing.
  • a target object processing method is provided.
  • the method is applied to the edge device side, and combines a first processing module for labeling first data and a second processing module for second data scene adaptation.
  • the processing module is deployed on the edge device side; the method includes: inputting the first data into the first processing module to obtain the predicted data labeling result; inputting the data labeling result to the second processing module , Performing scene adaptive incremental learning according to the data annotation result to obtain a neural network adapted to the second data scene; according to the data containing the target object and the neural network, realizing the scene corresponding to the target object Processing.
  • the method further includes: the scene is the current first scene, and performing adaptive incremental learning of the first scene according to the data labeling result to obtain the first scene suitable for the first scene.
  • the neural network After the neural network is configured, it is monitored that the scene is changed from the first scene to the second scene; the parameter resetting of the parameters in the second processing module is triggered; the second is performed according to the data annotation result
  • Scene adaptive incremental learning is used to obtain a neural network adapted to the second scene.
  • the method before the neural network is obtained, the method further includes: in the case of a parameter update in the second processing module, performing update restriction on the parameter update according to constraint conditions.
  • the inputting the first data into the first processing module to obtain the predicted data labeling result includes: obtaining a prediction rule according to the prediction of the scene data by the first processing module; Annotate the first data according to the prediction rule to obtain the predicted data annotation result.
  • the predicted data labeling result is derived from the first output of the output layer of the data sample training network constituting the first processing module; and/or, the predicted data labeling
  • the first output of the output layer of the training network and the second output of the intermediate layer are derived from the data samples constituting the first processing module.
  • the method further includes: before inputting the first data into the first processing module, selecting the first data according to sampling parameters to obtain the first data to be processed; Before obtaining the neural network, the method further includes:
  • the prediction results for the first data to be processed have a large difference
  • One or more frames of first data in the edge device perform the scene-adaptive incremental learning.
  • the scene adaptive incremental learning is performed on the edge device side for one or more frames of first data with a large difference in prediction results in the first data to be processed, including : Output one or more data labeling results obtained by the first processing module for the one or more frames of first data to the second processing module; according to the one or more data labeling results, the second The processing module performs training and updates the parameters in the second processing module;
  • the method further includes: before inputting the first data into the first processing module, selecting the first data according to sampling parameters to obtain the first data to be processed; Before obtaining the neural network, the method further includes: for the first data to be processed, when the difference between the prediction results respectively output by the first processing module and the second processing module is small, The pre-configured strategy reduces the value of the sampling parameter.
  • the method further includes: before inputting the first data into the first processing module, using part of the data in the first data as an online test set; and obtaining the neural network Previously, the method further includes: for the first data in the online test set, when the prediction results output by the first processing module and the second processing module differ significantly, the second processing module Parameter reset in the processing module.
  • the method further includes: after obtaining the neural network, monitoring edge devices in multiple regions to obtain a first edge device in an idle state; processing according to the edge device corresponding to the first edge device Ability to perform adaptive incremental training for the second data scene adaptation on the second processing module.
  • the method further includes: after obtaining the neural network, monitoring edge devices in multiple regions to obtain multiple edge device processing capabilities; according to the edge devices corresponding to the multiple edge devices respectively Processing capability and current resource consumption, select the second edge device with the highest processing capability of the edge device from the multiple edge devices; perform the second processing according to the processing capability of the edge device corresponding to the second edge device
  • the module performs adaptive incremental training adapted to the second data scene.
  • a target object processing device the device is deployed on the edge device side, the device includes a first processing module, a second processing module, and a third processing module; wherein the first processing module A processing module is configured to obtain the predicted data labeling result according to the input first data; the second processing module is configured to perform scene-adaptive incremental learning according to the input data labeling result to obtain the The neural network adapted to the second data scene; the third processing module is configured to process the scene corresponding to the target object according to the data containing the target object and the neural network.
  • the device further includes: a monitoring module configured to perform the adaptive incremental learning of the first scene according to the data annotation result, and the monitoring module is configured to After the neural network adapted to the first scene, it is monitored that the scene changes from the first scene to the second scene; a reset triggering module is configured to trigger the parameterization of the parameters in the second processing module Reset; the second processing module is further configured to perform the second scene adaptive incremental learning according to the data annotation result to obtain a neural network adapted to the second scene.
  • a monitoring module configured to perform the adaptive incremental learning of the first scene according to the data annotation result, and the monitoring module is configured to After the neural network adapted to the first scene, it is monitored that the scene changes from the first scene to the second scene
  • a reset triggering module is configured to trigger the parameterization of the parameters in the second processing module Reset
  • the second processing module is further configured to perform the second scene adaptive incremental learning according to the data annotation result to obtain a neural network adapted to the second scene.
  • the device further includes: a parameter update module configured to: in the case of parameter update, perform update restriction on the parameter update according to constraint conditions.
  • the first processing module is further configured to: obtain a prediction rule according to the prediction of the scene data; mark the first data according to the prediction rule to obtain the predicted Data annotation results.
  • the predicted data labeling result is derived from the first output of the output layer of the data sample training network constituting the first processing module; and/or, the predicted data labeling
  • the first output of the output layer of the training network and the second output of the intermediate layer are derived from the data samples constituting the first processing module.
  • the apparatus further includes: a fourth processing module configured to select the first data according to sampling parameters before inputting the first data into the first processing module, Obtain the first data to be processed; the device further includes: a fifth processing module configured to output the first data to be processed in the first processing module and the second processing module respectively In the case of a large difference in prediction results, the scene adaptive incremental learning is performed on the edge device side for one or more frames of first data with large prediction results in the first data to be processed.
  • the fifth processing module is further configured to: output one or more data labeling results obtained by the first processing module for the one or more frames of first data to The second processing module; training the second processing module according to one or more data annotation results and updating the parameters in the second processing module; increasing the value of the sampling parameter with a pre-configured strategy.
  • the device further includes: a sixth processing module configured to select the first data according to sampling parameters before inputting the first data into the first processing module, Obtain the first data to be processed; the device further includes: a seventh processing module configured to output the first data to be processed in the first processing module and the second processing module respectively In the case where the difference in prediction results is small, the value of the sampling parameter is reduced by a pre-configured strategy.
  • the device further includes: an eighth processing module configured to treat part of the first data as online data before inputting the first data into the first processing module Test set; the device further includes: a ninth processing module configured to: for the first data in the online test set, the difference between the prediction results respectively output by the first processing module and the second processing module If it is large, reset the parameters in the second processing module.
  • the apparatus further includes: a first device monitoring module configured to monitor edge devices in multiple areas after obtaining the neural network to obtain the first edge device in an idle state;
  • a training processing module is configured to perform adaptive incremental training of second data scene adaptation on the second processing module according to the processing capability of the edge device corresponding to the first edge device.
  • the device further includes: a second device monitoring module configured to monitor edge devices in multiple areas after obtaining the neural network to obtain processing capabilities of multiple edge devices; device selection The processing module is configured to select the second edge device with the highest processing capability of the edge device from the plurality of edge devices according to the processing capabilities of the edge devices and the current resource consumption respectively corresponding to the multiple edge devices; second The training processing module is configured to perform adaptive incremental training of second data scene adaptation on the second processing module according to the processing capability of the edge device corresponding to the second edge device.
  • an electronic device including: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured to execute the method described in any one of the above .
  • a computer storage medium that stores executable instructions in the computer storage medium, and the executable instructions implement the method described in any one of the foregoing when executed by a processor.
  • the first data is input into the first processing Module to obtain the predicted data annotation result; input the data annotation result into the second processing module, and perform scene adaptive incremental learning according to the data annotation result to obtain a neural network adapted to the second data scene
  • the processing of the scene corresponding to the target object is realized.
  • the scene adaptive incremental learning of the present disclosure is adopted to obtain a neural network adapted to the second data scene
  • the processing of the scene corresponding to the target object can be realized according to the data containing the target object and the neural network (for example, the target object is a human face).
  • the image detection processing of the human body or the face is realized), and the obtained training data is not much different from the data of the actual application scenario, which can meet the processing requirements of the application scenario and reduce the cost.
  • Fig. 1 is a first flowchart of a method for processing a target object according to an exemplary embodiment
  • Fig. 2 is a second flowchart of a method for processing a target object according to an exemplary embodiment
  • Fig. 3 is a third flowchart of a method for processing a target object according to an exemplary embodiment
  • Fig. 4 is a fourth flowchart of a method for processing a target object according to an exemplary embodiment
  • Fig. 5 is a block diagram of a device for processing a target object according to an exemplary embodiment
  • Fig. 6 is a second block diagram of a device for processing a target object according to an exemplary embodiment
  • Fig. 7 is a block diagram showing an electronic device 800 according to an exemplary embodiment
  • Fig. 8 is a block diagram showing an electronic device 900 according to an exemplary embodiment.
  • incremental learning refers to a model training program in which the model achieves better results on the new data during the continuous addition and training of new data.
  • Incremental learning methods are usually trained on two (or more) different data sets. Take two kinds of data sets as examples. First, train on data set A, and then use data set B for training. The focus is to achieve better prediction results on data set B while keeping basic data on data set A. The predictive effect of change.
  • Data-driven machine learning usually performs well when processing inputs that are similar to the training data. However, when the input and training data are far apart, due to the severe changes in the spatial relationship between its features and general features, the general model may Will perform poorly. However, collecting data and training for each application scenario is impractical or costly. However, the following embodiments of the present disclosure are adopted to achieve scene-adaptive incremental learning, and the scenario-adaptive incremental learning method is adopted.
  • an offline model (denoted as the T model) with higher accuracy but not satisfying the practicability is also designed.
  • the T model’s prediction of the application scenario data is used as an annotation, and the S model is incrementally trained on the edge device, so that the S model can be adapted to the application scenario, thereby obtaining better processing performance .
  • different learning strategies can be designed so that the S model can achieve the greatest performance improvement with as few iterations as possible.
  • the S model is reset to ensure the initial performance of the model, and then the processing performance of the model is trained and improved in the new application scenario.
  • Fig. 1 is a first flow chart of a method for processing a target object according to an exemplary embodiment.
  • the method for processing a target object is applied to a target object processing device, and the first processing module used for first data labeling and the second data
  • the second processing module for scene adaptation is deployed on the edge device side.
  • the target object processing device can be executed by a terminal device or a server or other processing device, where the terminal device can be a user equipment (UE, User Equipment), a mobile device, a cellular phone, a cordless phone, a personal digital processor (PDA, Personal Digital Assistant), handheld devices, computing devices, in-vehicle devices, wearable devices, etc.
  • the target object processing may be implemented by a processor invoking computer-readable instructions stored in the memory. As shown in Figure 1, the process includes:
  • Step S11 Input the first data into the first processing module to obtain the predicted data labeling result.
  • Step S12 Input the data annotation result into the second processing module, and perform scene adaptive incremental learning according to the data annotation result to obtain a neural network adapted to the second data scene.
  • Step S13 according to the data containing the target object and the neural network, realize the processing of the scene corresponding to the target object.
  • the first processing module and the second processing module can be obtained based on a calculation model.
  • the first processing module may be a T model
  • the second processing module may be an S model.
  • the first data is the original data obtained by collecting the target object.
  • the first data can include human body image data and face image data, or only any of human body image data and face image data A type of image data.
  • the T model can be an offline model
  • the S model can be an online model.
  • the second data is scene data, which is used for scene adaptation.
  • Application scenarios to which the present disclosure is applicable include: target detection, target recognition, example segmentation, super-resolution, reinforcement learning and other scenarios.
  • target detection in a surveillance scene the current smart cameras perform face or human detection on edge devices. Due to limited computing power, larger models cannot be used. Monitoring scenarios in different areas are very different, and the general model on the device cannot achieve good performance in various scenarios.
  • the position of the acquisition device such as a camera
  • the scene is solidified (the background is relatively single).
  • the problems that exist are the difficulty of scene adaptation, that is: on the one hand, training a model for each scene The cost is too high; on the other hand, when the location of the acquisition device (such as a camera) is fixed, the processing task of the target object (such as target detection) becomes simpler.
  • the processing flow in Figure 1 of the present disclosure you can use The computing power of edge devices in idle time, the model is incrementally trained for specific scenarios, thereby effectively improving task processing capabilities in specific scenarios (such as the detection rate of target detection).
  • the first data includes human body image data or human face image data.
  • the detection and processing of the image data can be completed independently without the edge device side and the cloud network. Since the entire process can be completed independently on the edge device side without the need for networking, the user's privacy data is protected.
  • Use the T model to predict the application scenario data to label the data eliminate the dependence on manual labeling data, and use the data labeling results for the data training of the S model, such as the incremental training on the edge device side, to achieve scene adaptation Incremental learning to achieve the purpose of adapting the S model to application scenarios and improving model performance.
  • the image data can be detected based on the data containing the target object and the neural network, and the obtained training data is consistent with the actual application scenario data There is not much difference, which can not only meet the detection requirements of application scenarios, but also reduce costs. There is no need to collect data and train for each application scenario, and it is suitable for all application scenarios.
  • Fig. 2 is a second flowchart of a target object processing method according to an exemplary embodiment.
  • the target object processing method is applied to a target object processing device.
  • the target object processing device can be executed by a terminal device or a server or other processing equipment.
  • the terminal device may be a user equipment (UE), a mobile device, a terminal, a cellular phone, a cordless phone, a personal digital processor (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc.
  • the target object processing may be implemented by a processor invoking computer-readable instructions stored in the memory.
  • the first processing module for labeling the first data and the second processing module for adapting to the second data scene are deployed on the edge device side. As shown in FIG. 2, the process includes:
  • Step S21 Input the first data into the first processing module to obtain the predicted data labeling result.
  • Step S22 The scene is the current first scene, and the first scene adaptive incremental learning is performed according to the data annotation result to obtain a neural network adapted to the first scene. According to the data containing the target object and the neural network, Realize the processing of the target object corresponding to the first scene.
  • Step S23 It is detected that the scene is changed from the first scene to the second scene, and parameter resetting of the parameters in the second processing module is triggered.
  • Step S24 Perform adaptive incremental learning of the second scene according to the data annotation result to obtain a neural network adapted to the second scene, and realize the processing of the target object corresponding to the second scene according to the data containing the target object and the neural network .
  • the first processing module and the second processing module can be obtained based on a calculation model.
  • the first processing module may be a T model
  • the second processing module may be an S model.
  • the first data may include human body image data and human face image data in the target detection scene, or may only include any one of human body image data and human face image data.
  • the T model can be an offline model
  • the S model can be an online model.
  • the T model and the S model can be two network models of different scales.
  • the T model can use a larger network structure to give it a strong predictive ability, while the S model is the actual model used in the product of the application scenario.
  • the T model is used for prediction in advance, and the obtained prediction result is used as an annotation for the training and learning of the small model.
  • the S model is reset to ensure the initial performance of the model, and then the processing performance of the model is trained and improved in the new application scenario.
  • Constraining the S model can also guarantee the initial performance of the S model, and then train and improve the processing performance of the model in new application scenarios.
  • the update of the parameter is restricted according to constraint conditions. For example, it can be achieved through a weight matrix. The purpose is to make the second processing module (S model) retain some initial state (when it is just deployed) during the change process, thereby retaining the characteristics of parameter solidification, and helping to avoid excessive attention to the current The scene causes problems such as over-fitting.
  • the inputting the first data into the first processing module to obtain the predicted data labeling result includes: obtaining a prediction rule according to the prediction of the scene data by the first processing module; Annotate the first data according to the prediction rule to obtain the predicted data annotation result.
  • the first processing module for the predicted data annotation result obtained by the first processing module, there are at least two sources as follows.
  • the first processing module can be understood as an annotation source.
  • the predicted data annotation result is derived from the first output of the output layer of the data sample training network constituting the first processing module.
  • the predicted data annotation result is derived from the data sample training network output layer that constitutes the first processing module The first output and the second output of the middle layer.
  • Fig. 3 is a third flowchart of a target object processing method according to an exemplary embodiment.
  • the target object processing method is applied to a target object processing apparatus.
  • the target object processing apparatus can be executed by a terminal device or a server or other processing equipment.
  • the terminal device may be a user equipment (UE), a mobile device, a cellular phone, a cordless phone, a personal digital processor (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc.
  • the target object processing may be implemented by a processor invoking computer-readable instructions stored in the memory. As shown in Figure 3, the process includes:
  • Step S31 Before inputting the first data into the first processing module, the first data is selected according to the sampling parameters to obtain the first data to be processed.
  • the sampling parameter is sampling frequency.
  • Step S32 For the first data to be processed, in the case where the prediction results output by the first processing module and the second processing module differ greatly, obtain the first data to be processed with a large difference in prediction results One or more frames of first data.
  • Step S33 Output one or more data labeling results obtained by the first processing module for the one or more frames of first data to the second processing module.
  • Step S34 Train the second processing module according to one or more data annotation results and update the parameters in the second processing module, and increase the value of the sampling parameter with a pre-configured strategy.
  • the scene adaptive incremental learning is performed on the edge device side to obtain the neural network adapted to the second data scene. This is only an optional implementation manner.
  • the first data is selected according to sampling parameters (such as sampling frequency) to obtain the first data to be processed.
  • sampling parameters such as sampling frequency
  • the method further includes: before inputting the first data into the first processing module, using part of the data in the first data as an online test set, and for the first data in the online test set, When the prediction results output by the first processing module and the second processing module differ significantly, the parameters in the second processing module are reset.
  • the method further includes: performing scene-adaptive incremental learning to obtain a neural network, and in the case of using the neural network to process a certain application scenario, the edges of multiple regions can be monitored
  • the device obtains the first edge device in an idle state, and performs adaptive incremental training of second data scene adaptation on the second processing module according to the processing capability of the edge device corresponding to the first edge device.
  • the method further includes: performing scene-adaptive incremental learning to obtain a neural network, and in the case of using the neural network to process a certain application scenario, the edges of multiple regions can be monitored
  • the device obtains processing capabilities of multiple edge devices, and selects a second edge device with high processing capability from the multiple edge devices according to their own processing capabilities and current resource consumption respectively corresponding to the multiple edge devices. According to the processing capability of the edge device corresponding to the second edge device, perform adaptive incremental training of the second data scene adaptation on the second processing module.
  • the model can be incrementally trained in specific scenarios, thereby effectively improving the detection rate of target object detection in specific scenarios.
  • Fig. 4 is a fourth flowchart of a target object processing method according to an exemplary embodiment.
  • the target object processing method is applied to a target object processing apparatus.
  • the target object processing apparatus may be executed by a terminal device or a server or other processing equipment.
  • the terminal device may be a user equipment (UE), a mobile device, a cellular phone, a cordless phone, a personal digital processor (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc.
  • the target object processing may be implemented by a processor invoking computer-readable instructions stored in the memory. As shown in Figure 4, the process includes:
  • the first processing module is the T model
  • the second processing module is the S model.
  • the S model and T model are deployed on the edge device side, and the edge device side is placed in a specific application scenario, such as a target detection scenario .
  • a specific application scenario such as a target detection scenario
  • the adaptive incremental learning of scene data is realized. For example, periodically select new data from a test set (such as an online test set), obtain a preset sampling frequency f, and select part of the image in the new data at the sampling frequency f (indicated by the dotted arrow in Figure 4), and select some
  • the image data of is sent to the S model and T model deployed on the edge device side for prediction. Compare the differences between the two models (S model and T model) through the model distillation evaluator.
  • the edge device side end training is performed on the image data with the large difference in the prediction results, that is, the S model is trained and the S model is output by the annotation result of the frame image of the T model.
  • the current parameter of the S model is ⁇ *.
  • the loss function also has a constraint model calculated by formula (1).
  • ⁇ represents the importance of the solidification weight, and ⁇ can be set to zero to give up the solidification model. If the prediction results output by the two models (S model and T model) differ slightly, the sampling frequency f is reduced or maintained. After the accumulation of time and training images, the S model will better adapt to the application scenario, that is, the output of the T model is used for the data training and learning of the S model, and the incremental learning of the scene adaptation is realized, so as to be adapted to the scene data According to the detection result, the target object is detected according to the detection result.
  • part of the selected image data is used as an online test set (indicated by a bold solid arrow in FIG. 4) in a preset manner, and the difference between the two models (S model and T model) is compared through the test evaluator. If the prediction results of the S model and the T model for the same online test set are quite different, the parameters ⁇ of the original S model can be reloaded to ensure model performance. If the application scene needs to be changed, the parameters ⁇ of the original S model can also be reloaded to ensure that the performance of the system in the new scene is not affected by the original scene.
  • the writing order of the steps does not mean a strict execution order but constitutes any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possibility.
  • the inner logic is determined.
  • the present disclosure also provides target object processing devices, electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any target object processing method provided in the present disclosure.
  • target object processing devices electronic equipment, computer-readable storage media, and programs, all of which can be used to implement any target object processing method provided in the present disclosure.
  • Fig. 5 is a block diagram 1 of a device for processing a target object according to an exemplary embodiment.
  • the device for processing a target object of an embodiment of the present disclosure is deployed on the edge device side, and the device includes a first processing module 201, a second processing module 202, and a third processing module 203; wherein the first processing module 201 is configured to obtain the predicted data labeling result according to the input first data.
  • the second processing module 202 is configured to perform scene adaptive incremental learning according to the input data annotation result to obtain a neural network adapted to the second data scene.
  • the third processing module 203 is configured to process the scene corresponding to the target object according to the data containing the target object and the neural network.
  • Fig. 6 is a second block diagram of a device for processing a target object according to an exemplary embodiment.
  • the device for processing a target object according to an embodiment of the present disclosure is deployed on the edge device side, and the device includes a first processing module 201, the second processing module 202, the third processing module 203, the monitoring module 204, and the reset triggering module 205; wherein the first processing module 201 is configured to obtain the predicted data labeling result according to the input first data.
  • the second processing module 202 is configured to perform scene adaptive incremental learning according to the input data annotation result to obtain a neural network adapted to the second data scene.
  • the third processing module 203 is configured to process the scene corresponding to the target object according to the data containing the target object and the neural network.
  • the monitoring module 204 is configured to perform the first-scene adaptive incremental learning according to the data labeling result, and obtain the neural network adapted to the first scene, and then monitor The scene is changed from the first scene to the second scene.
  • the reset triggering module 205 is configured to trigger parameter resetting of the parameters in the second processing module 202.
  • the second processing module 202 is further configured to perform the second scene adaptive incremental learning according to the data annotation result, to obtain a neural network adapted to the second scene.
  • the device further includes: a parameter update module 206 (not shown in FIG. 6), configured to: in the case of parameter update, update the parameter update according to the constraint condition.
  • a parameter update module 206 (not shown in FIG. 6), configured to: in the case of parameter update, update the parameter update according to the constraint condition.
  • the first processing module 201 is further configured to: obtain a prediction rule according to the prediction of the scene data; mark the first data according to the prediction rule to obtain the predicted data Mark the result.
  • the predicted data labeling result is derived from the first output of the output layer of the data sample training network constituting the first processing module; and/or the predicted data labeling result is derived from The data samples constituting the first processing module train the first output of the output layer of the network and the second output of the intermediate layer.
  • the device further includes: a fourth processing module 207 (not shown in FIG. 6), configured to perform processing of all data according to sampling parameters before inputting the first data into the first processing module 201 The first data is selected to obtain the first data to be processed.
  • the device also includes: a fifth processing module 208 (not shown in FIG. 6), configured to: for the first data to be processed, the prediction results respectively output by the first processing module 201 and the second processing module 202 In the case of a large difference, the scene adaptive incremental learning is performed on the edge device side for one or more frames of first data with large differences in prediction results in the first data to be processed.
  • the fifth processing module 208 is further configured to: output one or more data annotation results obtained by the first processing module 201 for the one or more frames of first data to the second Processing module 202.
  • the second processing module 202 is trained and the parameters in the second processing module 202 are updated, and the value of the sampling parameter is increased by a pre-configured strategy.
  • the device further includes: a sixth processing module 209 (not shown in FIG. 6), which is configured to perform processing on all data according to sampling parameters before inputting the first data into the first processing module 201 The first data is selected to obtain the first data to be processed.
  • the device also includes: a seventh processing module 210 (not shown in FIG. 6), configured to: for the first data to be processed, the prediction results respectively output by the first processing module 201 and the second processing module 202 When the difference is small, the value of the sampling parameter is reduced by a pre-configured strategy.
  • the device further includes: an eighth processing module 211 (not shown in FIG. 6), configured to input the first data into the first processing module 201 before inputting the first data Part of the data is used as an online test set.
  • the device further includes: a ninth processing module 212 (not shown in FIG. 6), configured to: for the first data in the online test set, the difference between the prediction results output by the first processing module 201 and the second processing module 202 In a larger case, the parameters in the second processing module 202 are reset.
  • the device further includes: a first device monitoring module 213 (not shown in FIG. 6), configured to monitor edge devices in multiple areas after obtaining the neural network, and obtain the State of the first edge device; the first training processing module 214 (not shown in FIG. 6), configured to perform the second data scene adaptation on the second processing module 202 according to the processing capability of the edge device corresponding to the first edge device Adaptive incremental training.
  • a first device monitoring module 213 (not shown in FIG. 6), configured to monitor edge devices in multiple areas after obtaining the neural network, and obtain the State of the first edge device
  • the first training processing module 214 (not shown in FIG. 6), configured to perform the second data scene adaptation on the second processing module 202 according to the processing capability of the edge device corresponding to the first edge device Adaptive incremental training.
  • the device further includes: a second device monitoring module 215 (not shown in FIG. 6), configured to monitor edge devices in multiple regions after obtaining the neural network, and obtain multiple Edge device processing capability; the device selection processing module 216 (not shown in FIG. 6) is configured to select from the multiple edge devices according to their own processing capabilities and current resource consumption respectively corresponding to the multiple edge devices The second edge device with high processing capability; the second training processing module 217 (not shown in FIG. 6) is configured to execute the second data on the second processing module 202 according to the processing capability of the edge device corresponding to the second edge device Adaptive incremental training for scene adaptation.
  • a second device monitoring module 215 configured to monitor edge devices in multiple regions after obtaining the neural network, and obtain multiple Edge device processing capability
  • the device selection processing module 216 (not shown in FIG. 6) is configured to select from the multiple edge devices according to their own processing capabilities and current resource consumption respectively corresponding to the multiple edge devices The second edge device with high processing capability
  • the second training processing module 217 (not shown in FIG. 6) is configured to execute
  • the functions or modules contained in the apparatus provided in the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments.
  • the functions or modules contained in the apparatus provided in the embodiments of the present disclosure can be used to execute the methods described in the above method embodiments.
  • the embodiments of the present disclosure also provide a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the foregoing method when executed by a processor.
  • the computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • An embodiment of the present disclosure also provides an electronic device, including: a processor; a memory for storing executable instructions of the processor; wherein the processor is configured as the above method.
  • the electronic device can be provided as a terminal, a server, or other forms of equipment.
  • Fig. 7 is a block diagram showing an electronic device 800 according to an exemplary embodiment.
  • the electronic device 800 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and other terminals.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power supply component 806, a multimedia component 808, an audio component 810, an input/output (I/O, Input/Output) interface 812 , The sensor component 814, and the communication component 816.
  • the processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
  • the processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method.
  • the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components.
  • the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
  • the memory 804 is configured to store various types of data to support operations in the electronic device 800. Examples of these data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc.
  • the memory 804 can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM, Static Random-Access Memory), electrically erasable programmable read-only memory (EEPROM). , Electrically-Erasable Programmable Read Only Memory, Erasable Programmable Read Only Memory (EPROM, Erasable Programmable Read Only Memory), Programmable Read-Only Memory (PROM, Programmable Read-only Memory), Read-Only Memory (ROM, Read Only Memory), magnetic memory, flash memory, magnetic disk or optical disk.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read-only memory
  • EPROM Erasable Programmable Read Only Memory
  • PROM Programmable Read-only Memory
  • ROM Read Only Memory
  • magnetic memory
  • the power supply component 806 provides power for various components of the electronic device 800.
  • the power supply component 806 may include a power management system, one or more power supplies, and other components associated with the generation, management, and distribution of power for the electronic device 800.
  • the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (LCD, Liquid Crystal Display) and a touch panel (TP, Touch Panel). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, sliding, 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 related to the touch or slide operation.
  • the multimedia component 808 includes a front camera and/or a rear camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (microphone, MIC for short), and when the electronic device 800 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive external audio signals.
  • the received audio signal may be further stored in the memory 804 or transmitted via the communication component 816.
  • the audio component 810 further includes a speaker for outputting audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 814 includes one or more sensors for providing the electronic device 800 with various aspects of state evaluation.
  • the sensor component 814 can detect the on/off status of the electronic device 800 and the relative positioning of the components.
  • the component is the display and the keypad of the electronic device 800.
  • the sensor component 814 can also detect the electronic device 800 or the electronic device 800.
  • the position of the component changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature change of the electronic device 800.
  • the sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
  • the sensor component 814 may also include a light sensor, such as a complementary metal oxide semiconductor (CMOS, Complementary Metal Oxide Semiconductor) or a charge-coupled device (CCD, Charge-coupled Device) image sensor, for use in imaging applications.
  • CMOS complementary metal oxide semiconductor
  • CCD Charge-coupled Device
  • the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor or a temperature sensor.
  • the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
  • the electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof.
  • the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a Near Field Communication (NFC, Near Field Communication) module to facilitate short-range communication.
  • NFC Near Field Communication
  • the NFC module can be based on radio frequency identification (RFID, Radio Frequency Identification) technology, infrared data association (IrDA, Infrared Data Association) technology, ultra-wideband (UWB, Ultra Wide Band) technology, Bluetooth (BT, Blue Tooth) technology and Other technologies to achieve.
  • RFID Radio Frequency Identification
  • IrDA Infrared Data Association
  • UWB Ultra Wide Band
  • Bluetooth Bluetooth
  • the electronic device 800 may be implemented by one or more application specific integrated circuits (ASIC, Application Specific Integrated Circuit), digital signal processor (DSP, Digital Signal Processor), and digital signal processing device (DSPD, Digital Signal Processor). Processing Device), Programmable Logic Device (PLD, Programmable Logic Device), Field Programmable Gate Array (FPGA, Field Programmable Gate Array), controller, microcontroller, microprocessor or other electronic components to implement the above method.
  • ASIC Application Specific Integrated Circuit
  • DSP Digital Signal Processor
  • DSPD Digital Signal Processor
  • Processing Device Programmable Logic Device
  • FPGA Field Programmable Gate Array
  • controller microcontroller, microprocessor or other electronic components to implement the above method.
  • a non-volatile computer-readable storage medium such as the memory 804 including computer program instructions, which can be executed by the processor 820 of the electronic device 800 to complete the foregoing method.
  • Fig. 8 is a block diagram showing an electronic device 900 according to an exemplary embodiment.
  • the electronic device 900 may be provided as a server.
  • the electronic device 900 includes a processing component 922, which further includes one or more processors, and a memory resource represented by a memory 932, for storing instructions that can be executed by the processing component 922, such as application programs.
  • the application program stored in the memory 932 may include one or more modules each corresponding to a set of instructions.
  • the processing component 922 is configured to execute instructions to perform the aforementioned methods.
  • the electronic device 900 may also include a power supply component 926 configured to perform power management of the electronic device 900, a wired or wireless network interface 950 configured to connect the electronic device 900 to a network, and an input output (I/O) interface 958 .
  • the electronic device 900 can operate based on an operating system stored in the memory 932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • a non-volatile computer-readable storage medium is also provided, such as a memory 932 including computer program instructions, which can be executed by the processing component 922 of the electronic device 900 to complete the foregoing method.
  • the present disclosure may be a system, method, and/or computer program product.
  • the computer program product may include a computer-readable storage medium loaded with computer-readable program instructions for enabling 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 instructions used by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • Non-exhaustive list of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM, Random Access Memory), read-only memory (ROM, Read Only Memory), erasable Programmable read-only memory (EPROM or flash memory), static random access memory (SRAM, Static Random-Access Memory), portable compact disc read-only memory (CD-ROM, Compact Disc-Read Only Memory), digital multi-function disk (DVD, Digital Versatile Disc), memory sticks, floppy disks, mechanical encoding devices, such as punch cards on which instructions are stored or raised structures in grooves, and any suitable combination of the above.
  • RAM random access memory
  • ROM read-only memory
  • EPROM or flash memory erasable Programmable read-only memory
  • SRAM Static Random-Access Memory
  • portable compact disc read-only memory CD-ROM, Compact Disc-Read Only Memory
  • DVD Digital Versatile Disc
  • memory sticks floppy disks
  • mechanical encoding devices such as punch cards on which instructions are stored or raised structures in grooves, and
  • the computer-readable storage medium used here is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (for example, light pulses through fiber optic cables), or through wires Transmission of electrical signals.
  • the computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing/processing devices, or downloaded 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, optical fiber 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 computer-readable program instructions from the network, and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device .
  • the computer program instructions used to perform the operations of the present disclosure may be assembly instructions, instruction set architecture (ISA, Instruction Set Architecture) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or one or more Source code or object code written in any combination of two programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as "C" language or similar programming languages.
  • Computer-readable program instructions can be executed entirely on the user's computer, partly on the user's computer, executed as a stand-alone software package, partly on the user's computer and partly executed on a remote computer, or entirely on the remote computer or server carried out.
  • the remote computer can be connected to the user's computer through any kind of network-including a local area network (LAN, Local Area Network) or a wide area network (WAN, Wide Area Network)-or it can be connected to an external computer (such as Use an Internet service provider to connect via the Internet).
  • the electronic circuit is customized by using the state information of the computer-readable program instructions, such as programmable logic circuit, Field-Programmable Gate Array (FPGA, Field-Programmable Gate Array), or Programmable Logic Array (PLA). , Programmable Logic Array), the electronic circuit can execute computer-readable program instructions to implement various aspects of the present disclosure.
  • These computer-readable program instructions can be provided to the processors of general-purpose computers, special-purpose computers, or other programmable data processing devices, thereby producing a machine that makes these instructions when executed by the processors of the computer or other programmable data processing devices , A device that implements the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams is produced. It is also possible to store these computer-readable program instructions in a computer-readable storage medium. These instructions make computers, programmable data processing apparatuses, and/or other devices work in a specific manner. Thus, the computer-readable medium storing the instructions includes An article of manufacture, which includes instructions for implementing various aspects of the functions/actions specified in one or more blocks in the flowchart and/or block diagram.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of an instruction, and the module, program segment, or part of an instruction contains one or more components for realizing the specified logical function.
  • Executable instructions may also occur in a different order from the order marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions Or it can be realized by a combination of dedicated hardware and computer instructions.
  • the first data is input to the first processing module to obtain the predicted data labeling result; the data labeling result is input to the second processing module, and the scene adaptive enhancement is performed according to the data labeling result.
  • the data containing the target object and the neural network realize the processing of the scene corresponding to the target object; in this way, according to the data containing the target object and the neural network
  • the network realizes the processing of the scene corresponding to the target object, and the obtained training data is not much different from the data of the actual application scene, which not only meets the processing requirements of the application scene, but also reduces the cost.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Electrical Discharge Machining, Electrochemical Machining, And Combined Machining (AREA)
  • Multi-Process Working Machines And Systems (AREA)
  • Supplying Of Containers To The Packaging Station (AREA)
PCT/CN2019/101448 2019-01-31 2019-08-19 一种目标对象处理方法、装置、电子设备及存储介质 Ceased WO2020155609A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
KR1020207018377A KR20200096556A (ko) 2019-01-31 2019-08-19 목표 대상 처리 방법, 장치, 전자 기기, 저장 매체
JP2020533136A JP7064593B2 (ja) 2019-01-31 2019-08-19 ターゲットオブジェクト処理方法、装置、電子機器及び記憶媒体
SG11202005886RA SG11202005886RA (en) 2019-01-31 2019-08-19 Target object processing method and apparatus, electronic device, and storage medium
US16/901,190 US11403489B2 (en) 2019-01-31 2020-06-15 Target object processing method and apparatus, electronic device, and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910098735.5 2019-01-31
CN201910098735.5A CN109766954B (zh) 2019-01-31 2019-01-31 一种目标对象处理方法、装置、电子设备及存储介质

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/901,190 Continuation US11403489B2 (en) 2019-01-31 2020-06-15 Target object processing method and apparatus, electronic device, and storage medium

Publications (1)

Publication Number Publication Date
WO2020155609A1 true WO2020155609A1 (zh) 2020-08-06

Family

ID=66455838

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/101448 Ceased WO2020155609A1 (zh) 2019-01-31 2019-08-19 一种目标对象处理方法、装置、电子设备及存储介质

Country Status (7)

Country Link
US (1) US11403489B2 (enExample)
JP (1) JP7064593B2 (enExample)
KR (1) KR20200096556A (enExample)
CN (1) CN109766954B (enExample)
SG (1) SG11202005886RA (enExample)
TW (1) TWI772668B (enExample)
WO (1) WO2020155609A1 (enExample)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112419413A (zh) * 2020-12-07 2021-02-26 萱闱(北京)生物科技有限公司 终端设备的运动方向监测方法、介质、装置和计算设备
CN115631363A (zh) * 2022-09-27 2023-01-20 北京航空航天大学杭州创新研究院 一种目标检测方法和系统
CN118820699A (zh) * 2024-09-13 2024-10-22 安徽至博光电科技股份有限公司 一种基于光纤传感的管道泄漏监测方法及系统

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11509542B2 (en) * 2018-05-02 2022-11-22 Telefonaktiebolaget Lm Ericsson (Publ) First network node, third network node, and methods performed thereby, for handling a performance of a radio access network
CN109766954B (zh) 2019-01-31 2020-12-04 北京市商汤科技开发有限公司 一种目标对象处理方法、装置、电子设备及存储介质
CN110147836B (zh) * 2019-05-13 2021-07-02 腾讯科技（深圳）有限公司 模型训练方法、装置、终端及存储介质
CN112417986B (zh) * 2020-10-30 2023-03-10 四川天翼网络股份有限公司 一种基于深度神经网络模型的半监督在线人脸识别方法及系统
US20220172064A1 (en) * 2020-12-02 2022-06-02 Htc Corporation Machine learning method and machine learning device for eliminating spurious correlation
CN113240119B (zh) * 2021-04-08 2024-03-19 南京大学 一种用于游戏ai策略解释的跨模型蒸馏装置
CN113205120B (zh) * 2021-04-20 2023-11-24 北京百度网讯科技有限公司 数据标注方法、装置、电子设备及可读存储介质
CN113111878B (zh) * 2021-04-30 2022-03-18 中北大学 一种复杂背景下的红外弱小目标检测方法
CN113364543B (zh) * 2021-06-03 2022-03-15 厦门大学 一种基于联邦强化学习的边缘计算模型训练方法
CN114572233B (zh) * 2022-03-25 2022-11-29 阿波罗智能技术(北京)有限公司 基于模型集合的预测方法、电子设备及自动驾驶车辆
US12495190B2 (en) * 2022-05-04 2025-12-09 International Business Machines Corporation Dynamically generated interactive video content
CN115797753A (zh) * 2022-11-01 2023-03-14 联想(北京)有限公司 一种处理方法、装置和电子设备
CN116050433B (zh) * 2023-02-13 2024-03-26 北京百度网讯科技有限公司 自然语言处理模型的场景适配方法、装置、设备及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160189048A1 (en) * 2014-12-30 2016-06-30 Winbond Electronics Corp. Data analysis system and method
CN106529485A (zh) * 2016-11-16 2017-03-22 北京旷视科技有限公司 用于获取训练数据的方法及装置
CN107944374A (zh) * 2017-11-20 2018-04-20 北京奇虎科技有限公司 视频数据中特定对象检测方法及装置、计算设备
CN109242013A (zh) * 2018-08-28 2019-01-18 北京九狐时代智能科技有限公司 一种数据标注方法、装置、电子设备及存储介质
CN109766954A (zh) * 2019-01-31 2019-05-17 北京市商汤科技开发有限公司 一种目标对象处理方法、装置、电子设备及存储介质

Family Cites Families (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7764231B1 (en) * 1996-09-09 2010-07-27 Tracbeam Llc Wireless location using multiple mobile station location techniques
TW200415524A (en) * 2002-10-24 2004-08-16 Univ Duke Binary prediction tree modeling with many predictors and its uses in clinical and genomic applications
US8582807B2 (en) * 2010-03-15 2013-11-12 Nec Laboratories America, Inc. Systems and methods for determining personal characteristics
CN103366163B (zh) * 2013-07-15 2016-07-20 北京丰华联合科技有限公司 基于增量学习的人脸检测系统和方法
US20160066894A1 (en) * 2014-03-21 2016-03-10 Razzberry Inc. Health state monitoring device
US10878320B2 (en) * 2015-07-22 2020-12-29 Qualcomm Incorporated Transfer learning in neural networks
CN106097346B (zh) * 2016-06-13 2019-08-27 中国科学技术大学 一种自学习的视频火灾探测方法
CN106919980B (zh) * 2017-01-24 2020-02-07 南京大学 一种基于神经节分化的增量式目标识别系统
US11017291B2 (en) * 2017-04-28 2021-05-25 Intel Corporation Training with adaptive runtime and precision profiling
CN107358257B (zh) * 2017-07-07 2019-07-16 华南理工大学 一种大数据场景下可增量学习的图像分类训练方法
CN108229591B (zh) * 2018-03-15 2020-09-22 北京市商汤科技开发有限公司 神经网络自适应训练方法和装置、设备、程序和存储介质
CN108764269A (zh) * 2018-04-03 2018-11-06 华南理工大学 一种基于时空约束增量学习的跨数据集行人再识别方法
CN108545556B (zh) * 2018-05-02 2019-10-01 中国科学院计算技术研究所 基于神经网络的信息处理装置及方法
CN108764487B (zh) * 2018-05-29 2022-07-08 北京百度网讯科技有限公司 用于生成模型的方法和装置、用于识别信息的方法和装置
CN109224442B (zh) * 2018-09-03 2021-06-11 腾讯科技（深圳）有限公司 虚拟场景的数据处理方法、装置及存储介质
US11158286B2 (en) * 2018-10-05 2021-10-26 Disney Enterprises, Inc. Machine learning color science conversion
CN113168541B (zh) * 2018-10-15 2024-07-09 泰立戴恩菲力尔商业系统公司 用于成像系统的深度学习推理系统和方法
US10867210B2 (en) * 2018-12-21 2020-12-15 Waymo Llc Neural networks for coarse- and fine-object classifications

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160189048A1 (en) * 2014-12-30 2016-06-30 Winbond Electronics Corp. Data analysis system and method
CN106529485A (zh) * 2016-11-16 2017-03-22 北京旷视科技有限公司 用于获取训练数据的方法及装置
CN107944374A (zh) * 2017-11-20 2018-04-20 北京奇虎科技有限公司 视频数据中特定对象检测方法及装置、计算设备
CN109242013A (zh) * 2018-08-28 2019-01-18 北京九狐时代智能科技有限公司 一种数据标注方法、装置、电子设备及存储介质
CN109766954A (zh) * 2019-01-31 2019-05-17 北京市商汤科技开发有限公司 一种目标对象处理方法、装置、电子设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112419413A (zh) * 2020-12-07 2021-02-26 萱闱(北京)生物科技有限公司 终端设备的运动方向监测方法、介质、装置和计算设备
CN112419413B (zh) * 2020-12-07 2024-01-05 萱闱(北京)生物科技有限公司 终端设备的运动方向监测方法、介质、装置和计算设备
CN115631363A (zh) * 2022-09-27 2023-01-20 北京航空航天大学杭州创新研究院 一种目标检测方法和系统
CN118820699A (zh) * 2024-09-13 2024-10-22 安徽至博光电科技股份有限公司 一种基于光纤传感的管道泄漏监测方法及系统

Also Published As

Publication number Publication date
TW202030648A (zh) 2020-08-16
US20200311476A1 (en) 2020-10-01
CN109766954B (zh) 2020-12-04
TWI772668B (zh) 2022-08-01
JP7064593B2 (ja) 2022-05-10
US11403489B2 (en) 2022-08-02
SG11202005886RA (en) 2020-11-27
JP2021514499A (ja) 2021-06-10
KR20200096556A (ko) 2020-08-12
CN109766954A (zh) 2019-05-17

Similar Documents

Publication Publication Date Title
WO2020155609A1 (zh) 一种目标对象处理方法、装置、电子设备及存储介质
TWI740309B (zh) 圖像處理方法及裝置、電子設備和電腦可讀儲存介質
CN109919300B (zh) 神经网络训练方法及装置以及图像处理方法及装置
WO2021008023A1 (zh) 图像处理方法及装置、电子设备和存储介质
WO2020133966A1 (zh) 锚点确定方法及装置、电子设备和存储介质
WO2021082241A1 (zh) 图像处理方法及装置、电子设备和存储介质
WO2021036382A1 (zh) 图像处理方法及装置、电子设备和存储介质
CN107784279B (zh) 目标跟踪方法及装置
CN113537048B (zh) 图像处理方法及装置、电子设备和存储介质
CN107341509B (zh) 卷积神经网络的训练方法、装置及可读存储介质
WO2021189704A1 (zh) 图像处理方法及装置、电子设备和存储介质
WO2021056770A1 (zh) 图像重建方法及装置、电子设备和存储介质
CN109165738B (zh) 神经网络模型的优化方法及装置、电子设备和存储介质
CN111242303A (zh) 网络训练方法及装置、图像处理方法及装置
WO2021169136A1 (zh) 图像处理方法及装置、电子设备和存储介质
CN111523555A (zh) 图像处理方法及装置、电子设备和存储介质
CN112115321B (zh) 内容推荐模型的训练方法及装置、电子设备和存储介质
CN109635926A (zh) 用于神经网络的注意力特征获取方法、装置及存储介质
CN109447258B (zh) 神经网络模型的优化方法及装置、电子设备和存储介质
CN113506324B (zh) 图像处理方法及装置、电子设备和存储介质
CN111988622B (zh) 视频预测方法及装置、电子设备和存储介质
CN110674416A (zh) 游戏推荐方法及装置
CN111860552A (zh) 基于核自编码器的模型训练方法、装置及存储介质
CN112800954A (zh) 文本检测方法及装置、电子设备和存储介质
CN112749709A (zh) 图像处理方法及装置、电子设备和存储介质

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2020533136

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20207018377

Country of ref document: KR

Kind code of ref document: A

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

Ref document number: 19913382

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19913382

Country of ref document: EP

Kind code of ref document: A1