CN111881764B - Target detection method and device, electronic equipment and storage medium - Google Patents
Target detection method and device, electronic equipment and storage medium Download PDFInfo
- Publication number
- CN111881764B CN111881764B CN202010626321.8A CN202010626321A CN111881764B CN 111881764 B CN111881764 B CN 111881764B CN 202010626321 A CN202010626321 A CN 202010626321A CN 111881764 B CN111881764 B CN 111881764B
- Authority
- CN
- China
- Prior art keywords
- target
- task
- loss
- loss function
- training
- 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.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 34
- 230000006870 function Effects 0.000 claims abstract description 106
- 238000012549 training Methods 0.000 claims abstract description 94
- 238000000034 method Methods 0.000 claims abstract description 66
- 238000013136 deep learning model Methods 0.000 claims abstract description 65
- 230000008569 process Effects 0.000 claims abstract description 39
- 238000010586 diagram Methods 0.000 description 8
- 230000008859 change Effects 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 2
- 238000010168 coupling process Methods 0.000 description 2
- 238000005859 coupling reaction Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000008685 targeting Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Abstract
The application discloses a target detection method, a target detection device, electronic equipment and a storage medium, wherein the target detection method comprises the following steps: obtaining a target sample to be detected; inputting a target sample to be detected into a trained deep learning model to obtain a target to be detected in the target sample to be detected, wherein the pre-trained deep model comprises a target classification task and a target position regression task, and the loss weight of a loss function of the target classification task and the loss weight of a loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model. Because the loss weights of the loss functions of the target classification task and the target position regression task are dynamically adjusted in the deep learning model training process, more proper loss weights of the loss functions of the target classification task and the target position regression task can be set in different target detection or different stages of the same target detection process.
Description
Technical Field
The present application relates to the field of computer vision, and in particular, to a target detection method, apparatus, electronic device, and storage medium.
Background
The task of target detection is to judge whether a target to be detected exists in one picture, and if the target to be detected exists, the position of the target is required to be output. Therefore, in the training stage of the target detection model, two kinds of loss functions of tasks are respectively used for judging whether positions of targets and regression targets exist in the picture by the measurement model, the two tasks are respectively called a target classification task and a target position regression task, and the corresponding loss functions are respectively a classification loss function and a regression loss function.
The weights in the existing classification and regression loss functions are fixed throughout the model training process, often set empirically, so that the task of specifically targeting a certain target may not be appropriate. Moreover, even for the same target detection task, the model training pre-stage and model training post-stage are different in bias for the classification task and the target regression task.
Disclosure of Invention
The application aims to provide a target detection method, a target detection device, electronic equipment and a storage medium, which can set more proper loss weights of a target classification task and a target position regression task in different stages of different target detection or the same target detection process.
According to a first aspect, in one embodiment, there is provided a target detection method, including:
obtaining a target sample to be detected;
inputting the target sample to be detected into a pre-trained deep learning model to obtain a target to be detected in the target sample to be detected; the pre-trained depth model comprises a target classification task and a target position regression task, wherein the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model.
Further, the loss weight of the loss function of the target classification task is obtained by dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps:
acquiring a training target sample set;
constructing a deep learning model, and inputting the training target sample set into the deep learning model;
acquiring the accuracy of the training target sample set in the current target classification task in the deep learning model training process;
and calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task.
Further, the loss weight of the loss function of the target position regression task is obtained by dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps:
acquiring the cross ratio of the training target sample set in the current target position regression task in the deep learning model training process; the intersection ratio is the ratio of the intersection of the target real position area and the target predicted position area to the union of the target real position area and the target predicted position area;
and calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task.
Further, the calculating the loss weight of the loss function of the current target classification task according to the accuracy rate of the training target sample set in the current target classification task comprises:
the loss weight of the loss function of the current target classification task is calculated according to the following formula:
FL1＝-(1-A) γ log(A)
wherein FL1 is the loss weight of the loss function of the current target classification task, A is the accuracy of the training target sample set in the current target classification task, and gamma is a constant.
Further, the calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task comprises:
the loss weight of the loss function of the current target position regression task is calculated according to the following formula:
FL2＝-(1-IoU) γ log(IoU)
wherein FL2 is the loss weight of the loss function of the current target position regression task, ioU is the cross ratio of the training target sample set in the current target position regression task, and gamma is a constant.
Further, the loss function of the target position regression task is a loss function for the target position regression task in the SSD algorithm, and the loss function of the target classification task is a loss function for the target classification task in the SSD algorithm.
Further, the target sample is a face picture, and the target to be detected is a face.
According to a second aspect, there is provided in one embodiment an object detection apparatus comprising:
the sample acquisition module is used for acquiring a target sample to be detected and a training target sample set;
the target detection module is used for inputting the target sample to be detected into a pre-trained deep learning model to obtain a target to be detected in the target sample to be detected; the pre-trained depth model comprises a target classification task and a target position regression task, wherein the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model.
According to a third aspect, an embodiment provides an electronic device, including:
a memory for storing a program;
and a processor, configured to implement the method according to the above embodiment by executing the program stored in the memory.
According to a fourth aspect, an embodiment provides a computer readable storage medium comprising a program executable by a processor to implement a method as described in the above embodiments.
According to the target detection method, the device, the electronic equipment and the storage medium, the loss weights of the loss functions of the target classification tasks and the loss weights of the loss functions of the target position regression tasks are dynamically adjusted in the training process of the deep learning model, so that more proper loss weights of the loss functions of the target classification tasks and the target position regression tasks can be set in different target detection or different stages of the same target detection process.
Drawings
FIG. 1 is a flow chart of a target detection method of the present application;
FIG. 2 is a flow chart of a method of object detection according to one embodiment;
FIG. 3 is a schematic diagram of the intersection of a real frame and a predicted frame of a target, wherein (a) is a schematic diagram of the intersection of a real frame and a predicted frame, and (b) is a schematic diagram of the intersection of a real frame and a predicted frame;
FIG. 4 is a block diagram of an object detection device according to an embodiment;
fig. 5 is a block diagram of an electronic device of an embodiment.
Detailed Description
The application will be described in further detail below with reference to the drawings by means of specific embodiments. Wherein like elements in different embodiments are numbered alike in association. In the following embodiments, numerous specific details are set forth in order to provide a better understanding of the present application. However, one skilled in the art will readily recognize that some of the features may be omitted, or replaced by other elements, materials, or methods in different situations. In some instances, related operations of the present application have not been shown or described in the specification in order to avoid obscuring the core portions of the present application, and may be unnecessary to persons skilled in the art from a detailed description of the related operations, which may be presented in the description and general knowledge of one skilled in the art.
Furthermore, the described features, operations, or characteristics of the description may be combined in any suitable manner in various embodiments. Also, various steps or acts in the method descriptions may be interchanged or modified in a manner apparent to those of ordinary skill in the art. Thus, the various orders in the description and drawings are for clarity of description of only certain embodiments, and are not meant to be required orders unless otherwise indicated.
The numbering of the components itself, e.g. "first", "second", etc., is used herein merely to distinguish between the described objects and does not have any sequential or technical meaning. The term "coupled" as used herein includes both direct and indirect coupling (coupling), unless otherwise indicated.
In the embodiment of the application, in the process of training the deep learning model for detecting the target, the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task are dynamically adjusted, so that the loss weights of the loss functions of the target classification task and the target position regression task can be set more appropriately, and the result of the trained deep learning model for detecting the target is more accurate.
Referring to fig. 1, fig. 1 is a flowchart of a target detection method, which includes steps S10 to S20.
Step S10, a target sample to be detected is obtained.
Step S20, inputting a target sample to be detected into a pre-trained deep learning model to obtain a target to be detected in the target sample to be detected; the pre-trained depth model comprises a target classification task and a target position regression task, wherein the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model.
Referring to fig. 2, fig. 2 is a flowchart of an object detection method according to an embodiment, in which a face picture is taken as an example, the method is described, and the method includes steps S101 to S105.
Step S101, a training target sample set is acquired. In this embodiment, the training sample set is a picture having a face tag, where the face tag is used to identify a face and a position of the face in the picture, where the position of the face in the picture is represented by a real frame including the face. The method for training the face set obtained in this step may be a method commonly used in the prior art, for example, a method for monitoring face images in videos captured by a camera.
Step S102, a deep learning model is built, and a loss function of a target classification task and a target position regression task in the deep learning model is set. And taking the face pictures in the training target sample set as input, taking face features corresponding to the face pictures as labels, performing machine learning, and training to obtain a model function of the corresponding relation between the face pictures and the face features, wherein the model function is a deep learning model. In the training process, the deep learning model detects the target by continuously adjusting the position of the prediction frame in the picture, if the prediction frame is overlapped with the real frame in the picture, the detection is completed, the deep learning model comprises a cascade target classification task and a target position regression task, the target classification task is used for judging whether the target exists in the prediction frame or not, the target classification task comprises the existence or nonexistence of two results, and the target position regression task is used for judging whether the prediction frame is overlapped with the real frame or not, namely, the accurate position of the target is detected.
In this embodiment, the loss functions of the target classification task and the target position regression task in the deep learning model may adopt classification loss functions and regression loss functions in the SSD algorithm.
Wherein the classification loss function is:L 1 for the classification loss value, x is the scoreClass probability values.
The regression loss function is:L 2 for regression loss value, y is the position of the real box, +.>Is the position of the prediction box.
Step S103, inputting a training target sample set into the constructed deep learning model for training, and dynamically adjusting the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task in the deep learning model training process.
In step S102, the loss weights of the classification loss function and the regression loss function in the SSD algorithm are both 1, and since the problem that the first half deep learning model in the training process has a target in judging the sample is large, the loss weight of the classification loss function needs to be increased; the problem of judging whether the sample has the target is not great in the second-half deep learning model of the training process, and how to more accurately position the target is more critical, and the loss weight of the regression loss function should be increased at this time, so the embodiment dynamically adjusts the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task according to different stages.
Wherein dynamically adjusting the loss weight of the loss function of the target classification task during the deep learning model training process comprises: acquiring the accuracy of a training target sample set in a current target classification task in the deep learning model training process; and calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task. Because all training target samples in the training target sample set are provided with the face labels, the classification accuracy can be determined according to the face labels and the classification results output in the current target classification task, and the accuracy in different training periods is changed, for example, the classification accuracy is very low in the initial training period, and the classification accuracy is higher when the training period is reached to the later training period.
The present embodiment calculates the loss weight of the loss function of the current target classification task according to the following formula:
FL1＝-(1-A) γ log(A)
wherein FL1 is the loss weight of the loss function of the current target classification task, A is the accuracy of the target sample set in the current target classification task, gamma is a constant, and the value of gamma is 0-5.
Similarly, dynamically adjusting the loss weight of the loss function of the target position regression task in the deep learning model training process comprises: acquiring the cross ratio of a training target sample set in a current target position regression task in the deep learning model training process; the intersection ratio is the ratio of the intersection of the target real position area and the target predicted position area to the union of the target real position area and the target predicted position area; and calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the target sample set in the current target position regression task.
In this embodiment, the target real position area is the target real frame 301, the target predicted position area is the target predicted frame 302, as shown in fig. 3, fig. 3 (a) is a schematic diagram of the intersection of the real frame 301 and the predicted frame 302, and fig. 3 (b) is a schematic diagram of the union of the real frame 301 and the predicted frame 302. Wherein the intersection ratio is the ratio of the intersection area of the real frame 301 and the predicted frame 302 to the union area of the real frame 301 and the predicted frame 302.
The present embodiment calculates the loss weight of the loss function of the current target position regression task according to the following formula:
FL2＝-(1-IoU) γ log(IoU)
wherein FL2 is the loss weight of the loss function of the current target position regression task, ioU is the cross ratio of the target sample set in the current target position regression task, gamma is a constant, and the value of gamma is 0-5.
Loss weights of the loss functions of the current target classification task and the target position regression task calculated based on the embodiment, then in the SDD algorithm in the embodimentClassification loss function and regression loss function of FL1 x L, respectively 1 And FL2 x L 2 In the training process of the deep learning model, FL1 and FL2 change along with the change of the accuracy A of the target classification task and the intersection ratio IoU of the target position regression task, so that the classification loss function and the regression loss function also change, and the deep learning model is trained more efficiently and accurately.
Step S104, obtaining a target sample to be detected. The target samples in this embodiment are face images, and the target sample to be detected is an image of whether a face exists or not and the position of the face to be detected.
Step S105, inputting the target sample to be detected into the trained deep learning model to obtain the target to be detected in the target sample to be detected. After the training of the deep learning model is completed, the target sample to be detected is input into the trained deep learning model, and then the target in the target sample to be detected, for example, the face in the face picture can be output. According to the embodiment, the target to be detected can be extracted from the target sample to be detected based on the detected target to be detected, the extracted target to be detected is compared with the preset target in similarity, and if the comparison result shows that the target to be detected is similar to the preset target, the target to be detected is marked as a label of the preset target, so that the track of the preset target can be tracked conveniently.
In this embodiment, when the deep learning model just starts training, the target classification task at this time is relatively inaccurate, so the accuracy a of the target classification task is low, the loss weight FL1 of the loss function of the current target classification task is large, and the deep learning model tends to optimize the target classification task preferentially; when the deep learning model is trained to a later stage, the target classification task is accurate, and the target position regression task is inaccurate, so that the intersection ratio of the target sample set in the current target position regression task is relatively lower than IoU, the loss weight of the loss function of the current target position regression task is larger than FL2, the loss weight of the target position regression task is increased, the deep learning model tends to optimize the target position regression task, and the loss weights of the target classification task and the target position regression task are dynamically adjusted.
Referring to fig. 4, fig. 4 is a block diagram of an object detection device according to an embodiment, where the object detection device includes: a sample acquisition module 101 and a target detection module 102.
The sample acquisition module 101 is configured to acquire a target sample to be detected.
The target detection module 102 is configured to input a target sample to be detected into a pre-trained deep learning model, so as to obtain a target to be detected in the target sample to be detected; the pre-trained depth model comprises a target classification task and a target position regression task, wherein the loss weight of the loss function of the target classification task and the loss weight of the loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model.
The method comprises the steps of obtaining the loss weight of the loss function of the target classification task through dynamic adjustment in the training process of the deep learning model, wherein the method comprises the following steps: acquiring a training target sample set; constructing a deep learning model, and inputting a training target sample set into the deep learning model; acquiring the accuracy of a training target sample set in a current target classification task in the deep learning model training process; and calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task. The loss weight of the loss function of the target position regression task is obtained by dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps: acquiring the cross ratio of a training target sample set in a current target position regression task in the deep learning model training process; the intersection ratio is the ratio of the intersection of the target real position area and the target predicted position area to the union of the target real position area and the target predicted position area; and calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task.
The functions implemented by each module in the apparatus of this embodiment correspond to the steps in the method of the foregoing embodiment, and specific implementation and technical effects thereof refer to descriptions of the steps of the method of the foregoing embodiment, which are not repeated herein.
Referring to fig. 5, an embodiment of the present application provides an electronic device. The electronic device comprises, among other things, a memory 201, a processor 202, an input/output interface 203. Wherein the memory 201 is used for storing a program. The processor 202 is configured to invoke the program stored in the memory 201 to execute the target detection method according to the embodiment of the present application. The processor 202 is connected to the memory 201, the input/output interface 203, respectively, such as via a bus system and/or other form of connection mechanism (not shown). The memory 201 may be used to store programs and data, including feature fusion programs involved in embodiments of the present application, and the processor 202 performs various functional applications of the electronic device and data processing by running the programs stored in the memory 201.
Those skilled in the art will appreciate that all or part of the functions of the various methods in the above embodiments may be implemented by hardware, or may be implemented by a computer program. When all or part of the functions in the above embodiments are implemented by means of a computer program, the program may be stored in a computer readable storage medium, and the storage medium may include: read-only memory, random access memory, magnetic disk, optical disk, hard disk, etc., and the program is executed by a computer to realize the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the functions described above can be realized. In addition, when all or part of the functions in the above embodiments are implemented by means of a computer program, the program may be stored in a storage medium such as a server, another computer, a magnetic disk, an optical disk, a flash disk, or a removable hard disk, and the program in the above embodiments may be implemented by downloading or copying the program into a memory of a local device or updating a version of a system of the local device, and when the program in the memory is executed by a processor.
The foregoing description of the application has been presented for purposes of illustration and description, and is not intended to be limiting. Several simple deductions, modifications or substitutions may also be made by a person skilled in the art to which the application pertains, based on the idea of the application.
Claims (6)
1. A method of detecting an object, comprising:
obtaining a target sample to be detected;
inputting the target sample to be detected into a pre-trained deep learning model to obtain a target to be detected in the target sample to be detected; the pre-trained depth model comprises a target classification task and a target position regression task, wherein the loss weight of a loss function of the target classification task and the loss weight of a loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model;
the loss weight of the loss function of the target classification task is obtained by dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps:
acquiring a training target sample set;
constructing a deep learning model, and inputting the training target sample set into the deep learning model;
acquiring the accuracy of the training target sample set in the current target classification task in the deep learning model training process;
calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task;
the loss weight of the loss function of the target position regression task is obtained through dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps:
acquiring the cross ratio of the training target sample set in the current target position regression task in the deep learning model training process; the intersection ratio is the ratio of the intersection of the target real position area and the target predicted position area to the union of the target real position area and the target predicted position area;
calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task;
the calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task comprises the following steps:
the loss weight of the loss function of the current target classification task is calculated according to the following formula:
FL1＝-(1-A) γ log(A)
wherein FL1 is the loss weight of the loss function of the current target classification task, A is the accuracy of the training target sample set in the current target classification task, gamma is a constant, and the value of gamma is 0-5;
the calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task comprises the following steps:
the loss weight of the loss function of the current target position regression task is calculated according to the following formula:
FL2＝-(1-IoU) γ log(IoU)
wherein FL2 is the loss weight of the loss function of the current target position regression task, ioU is the cross ratio of the training target sample set in the current target position regression task, gamma is a constant, and the value of gamma is 0-5.
2. The method of claim 1, wherein the loss function of the destination location regression task is a loss function for the destination location regression task in an SSD algorithm, and the loss function of the destination classification task is a loss function for the destination classification task in the SSD algorithm.
3. The method for detecting an object according to claim 1, wherein the object sample is a face picture and the object to be detected is a face.
4. An object detection apparatus, comprising:
the sample acquisition module is used for acquiring a target sample to be detected;
the target detection module is used for inputting the target sample to be detected into a pre-trained deep learning model to obtain a target to be detected in the target sample to be detected; the pre-trained depth model comprises a target classification task and a target position regression task, wherein the loss weight of a loss function of the target classification task and the loss weight of a loss function of the target position regression task are obtained through dynamic adjustment in the training process of the deep learning model;
the loss weight of the loss function of the target classification task is obtained by dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps:
acquiring a training target sample set;
constructing a deep learning model, and inputting the training target sample set into the deep learning model;
acquiring the accuracy of the training target sample set in the current target classification task in the deep learning model training process;
calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task;
the loss weight of the loss function of the target position regression task is obtained through dynamic adjustment in the training process of the deep learning model, and the method comprises the following steps:
acquiring the cross ratio of the training target sample set in the current target position regression task in the deep learning model training process; the intersection ratio is the ratio of the intersection of the target real position area and the target predicted position area to the union of the target real position area and the target predicted position area;
calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task;
the calculating the loss weight of the loss function of the current target classification task according to the accuracy of the training target sample set in the current target classification task comprises the following steps:
the loss weight of the loss function of the current target classification task is calculated according to the following formula:
FL1＝-(1-A) γ log(A)
wherein FL1 is the loss weight of the loss function of the current target classification task, A is the accuracy of the training target sample set in the current target classification task, gamma is a constant, and the value of gamma is 0-5;
the calculating the loss weight of the loss function of the current target position regression task according to the intersection ratio of the training target sample set in the current target position regression task comprises the following steps:
the loss weight of the loss function of the current target position regression task is calculated according to the following formula:
FL2＝-(1-IoU) γ log(IoU)
wherein FL2 is the loss weight of the loss function of the current target position regression task, ioU is the cross ratio of the training target sample set in the current target position regression task, gamma is a constant, and the value of gamma is 0-5.
5. An electronic device, comprising:
a memory for storing a program;
a processor for implementing the method of any one of claims 1-3 by executing a program stored in the memory.
6. A computer readable storage medium comprising a program executable by a processor to implement the method of any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010626321.8A CN111881764B (en) | 2020-07-01 | 2020-07-01 | Target detection method and device, electronic equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010626321.8A CN111881764B (en) | 2020-07-01 | 2020-07-01 | Target detection method and device, electronic equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111881764A CN111881764A (en) | 2020-11-03 |
| CN111881764B true CN111881764B (en) | 2023-11-03 |
Family
ID=73150098
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010626321.8A Active CN111881764B (en) | 2020-07-01 | 2020-07-01 | Target detection method and device, electronic equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111881764B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112509008B (en) * | 2020-12-15 | 2022-05-03 | 重庆邮电大学 | Target tracking method based on cross-over ratio guided twin network |
| CN112613462B (en) * | 2020-12-29 | 2022-09-23 | 安徽大学 | Weighted intersection ratio method |
| CN113011597B (en) * | 2021-03-12 | 2023-02-28 | 山东英信计算机技术有限公司 | Deep learning method and device for regression task |
| CN113469025A (en) * | 2021-06-29 | 2021-10-01 | 阿波罗智联(北京)科技有限公司 | Target detection method and device applied to vehicle-road cooperation, road side equipment and vehicle |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106910013A (en) * | 2017-02-16 | 2017-06-30 | 中国科学院自动化研究所 | Unreal information detecting method and device based on Expression study |
| CN109272060A (en) * | 2018-09-06 | 2019-01-25 | 湖北工业大学 | A kind of method and system carrying out target detection based on improved darknet neural network |
| CN109858486A (en) * | 2019-01-27 | 2019-06-07 | 中国人民解放军国防科技大学 | Deep learning-based data center cloud target identification method |
| CN109886357A (en) * | 2019-03-13 | 2019-06-14 | 哈尔滨工程大学 | A kind of adaptive weighting deep learning objective classification method based on Fusion Features |
| CN110555390A (en) * | 2019-08-09 | 2019-12-10 | 厦门市美亚柏科信息股份有限公司 | pedestrian re-identification method, device and medium based on semi-supervised training mode |
| CN110827253A (en) * | 2019-10-30 | 2020-02-21 | 北京达佳互联信息技术有限公司 | Training method and device of target detection model and electronic equipment |
| CN111091105A (en) * | 2019-12-23 | 2020-05-01 | 郑州轻工业大学 | Remote sensing image target detection method based on new frame regression loss function |
| CN111310826A (en) * | 2020-02-13 | 2020-06-19 | 南京旷云科技有限公司 | Method and device for detecting labeling abnormity of sample set and electronic equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7769228B2 (en) * | 2004-05-10 | 2010-08-03 | Siemens Corporation | Method for combining boosted classifiers for efficient multi-class object detection |
| US10515296B2 (en) * | 2017-11-14 | 2019-12-24 | Adobe Inc. | Font recognition by dynamically weighting multiple deep learning neural networks |
-
2020
- 2020-07-01 CN CN202010626321.8A patent/CN111881764B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106910013A (en) * | 2017-02-16 | 2017-06-30 | 中国科学院自动化研究所 | Unreal information detecting method and device based on Expression study |
| CN109272060A (en) * | 2018-09-06 | 2019-01-25 | 湖北工业大学 | A kind of method and system carrying out target detection based on improved darknet neural network |
| CN109858486A (en) * | 2019-01-27 | 2019-06-07 | 中国人民解放军国防科技大学 | Deep learning-based data center cloud target identification method |
| CN109886357A (en) * | 2019-03-13 | 2019-06-14 | 哈尔滨工程大学 | A kind of adaptive weighting deep learning objective classification method based on Fusion Features |
| CN110555390A (en) * | 2019-08-09 | 2019-12-10 | 厦门市美亚柏科信息股份有限公司 | pedestrian re-identification method, device and medium based on semi-supervised training mode |
| CN110827253A (en) * | 2019-10-30 | 2020-02-21 | 北京达佳互联信息技术有限公司 | Training method and device of target detection model and electronic equipment |
| CN111091105A (en) * | 2019-12-23 | 2020-05-01 | 郑州轻工业大学 | Remote sensing image target detection method based on new frame regression loss function |
| CN111310826A (en) * | 2020-02-13 | 2020-06-19 | 南京旷云科技有限公司 | Method and device for detecting labeling abnormity of sample set and electronic equipment |
Non-Patent Citations (3)
| Title |
|---|
| Dynamic Saliency-Aware Regularization for Correlation Filter-Based Object Tracking;Wei Feng等;《IEEE TRANSACTIONS ON IMAGE PROCESSING》;第28卷(第7期);第3232-3245页 * |
| 基于联合优化的强耦合孪生区域推荐网络的目标跟踪算法;石国强 等;《计算机应用》;第40卷(第10期);第2822-2830页 * |
| 物联网安全技术分析;谭喆;《中国安防》(第10期);第67-73页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111881764A (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111881764B (en) | Target detection method and device, electronic equipment and storage medium | |
| KR102382693B1 (en) | Learning method and learning device of pedestrian detector for robust surveillance based on image analysis by using gan and testing method and testing device using the same | |
| US6226388B1 (en) | Method and apparatus for object tracking for automatic controls in video devices | |
| CN110569703B (en) | Computer-implemented method and device for identifying damage from picture | |
| US20020067857A1 (en) | System and method for classification of images and videos | |
| TW201832158A (en) | Determining recommended object | |
| US20200242333A1 (en) | Generating object embeddings from images | |
| US11386710B2 (en) | Eye state detection method, electronic device, detecting apparatus and computer readable storage medium | |
| CN110737785B (en) | Picture labeling method and device | |
| CN111079638A (en) | Target detection model training method, device and medium based on convolutional neural network | |
| EP3905122A2 (en) | Video type detection method, apparatus, electronic device and storage medium | |
| CN113129335B (en) | Visual tracking algorithm and multi-template updating strategy based on twin network | |
| US20220067454A1 (en) | Automatic work order analysis | |
| JP2019091121A (en) | Information processing device, background update method and background update program | |
| Minkina et al. | Generalization of the Viola-Jones method as a decision tree of strong classifiers for real-time object recognition in video stream | |
| CN114463603B (en) | Training method and device for image detection model, electronic equipment and storage medium | |
| CN113592706A (en) | Method and device for adjusting homography matrix parameters | |
| US8249354B2 (en) | System and method for finding edge points of an object | |
| CN110210314B (en) | Face detection method, device, computer equipment and storage medium | |
| Francis et al. | Pre-processing techniques for detection of blurred images | |
| CN115994899A (en) | Bolt loosening detection method, device and detection equipment | |
| CN113869163B (en) | Target tracking method and device, electronic equipment and storage medium | |
| CN112418217A (en) | Method, apparatus, device and medium for recognizing characters | |
| CN112561956A (en) | Video target tracking method and device, electronic equipment and storage medium | |
| CN112131418A (en) | Target labeling method, target labeling device and computer-readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |