CN113449539A - Training method, device, equipment and storage medium for animal body information extraction model - Google Patents
Training method, device, equipment and storage medium for animal body information extraction model Download PDFInfo
- Publication number
- CN113449539A CN113449539A CN202010211682.6A CN202010211682A CN113449539A CN 113449539 A CN113449539 A CN 113449539A CN 202010211682 A CN202010211682 A CN 202010211682A CN 113449539 A CN113449539 A CN 113449539A
- Authority
- CN
- China
- Prior art keywords
- model
- animal body
- training
- target
- loss value
- 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.)
- Withdrawn
Links
- 238000012549 training Methods 0.000 title claims abstract description 153
- 238000000605 extraction Methods 0.000 title claims abstract description 146
- 238000000034 method Methods 0.000 title claims abstract description 63
- 230000011218 segmentation Effects 0.000 claims abstract description 71
- 238000005259 measurement Methods 0.000 claims description 34
- 238000004590 computer program Methods 0.000 claims description 14
- 210000001364 upper extremity Anatomy 0.000 claims description 10
- 238000001514 detection method Methods 0.000 abstract description 32
- 230000006870 function Effects 0.000 description 36
- 230000037396 body weight Effects 0.000 description 19
- 238000012545 processing Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 4
- 241001083912 Cotyledon orbiculata Species 0.000 description 3
- 235000013146 Solanum mammosum Nutrition 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 210000005069 ears Anatomy 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 210000000746 body region Anatomy 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Classifications
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the application discloses a training method, a device, equipment and a storage medium for an animal information extraction model. The training method of the animal body information extraction model comprises the following steps: acquiring a training sample set, wherein the training sample set comprises sample images of a plurality of animal bodies, and the sample images are marked with contour labels and a plurality of key point labels of the animal bodies; determining a segmentation loss value and a key point loss value of the model to be trained according to the sample image; determining a total loss value according to the segmentation loss value and the key point loss value; and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model. According to the embodiment of the application, the detection accuracy of the key point data and the segmentation data of the animal body can be improved.
Description
Technical Field
The embodiment of the application relates to the technical field of computer vision, in particular to a training method, a device, equipment and a storage medium for an animal information extraction model.
Background
In the technical field of computer vision, it is often necessary to identify key points of an object through an image and detect segmentation data of the object through the image so as to further perform data processing for other purposes according to key point information and segmentation data of the object; for example, in order to measure the size information of the pig body, it is generally necessary to detect the segmentation data of the pig body and identify key points of the pig body (e.g., the pig's mouth, the pig's tail, the pig's ears, the pig's front legs, the pig's rear legs, etc.).
In the prior art, a sample image and manually labeled key points are usually used as training data sets directly to train a model for detecting key points of an object, so as to detect the key points of the object by using the trained key point detection model. Or directly using the sample image and the manually marked segmentation data as a training data set to train a model for detecting the object-related segmentation data, so as to detect the segmentation data of the object by adopting the trained segmentation data detection model. However, in practical applications, it is found that the existing single key point detection model and the single segmentation data detection model are not high in accuracy.
Disclosure of Invention
The embodiment of the application provides a training method, a training device, equipment and a storage medium for an animal body information extraction model, which are used for improving the accuracy of a key point detection result of the model to a certain extent and improving the detection accuracy of the model on segmentation data.
In a first aspect, an embodiment of the present application provides a method for training an animal body information extraction model, where the method includes:
acquiring a training sample set, wherein the training sample set comprises sample images of a plurality of animal bodies, and the sample images are marked with contour labels and a plurality of key point labels of the animal bodies;
determining a segmentation loss value and a key point loss value of the model to be trained according to the sample image;
determining a total loss value according to the segmentation loss value and the key point loss value;
and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
In some embodiments of the present application, the method further comprises:
acquiring an image to be detected, and inputting the image to be detected into the animal information extraction model;
and acquiring a segmentation map and target key points of a target foreground output by the animal information extraction model according to the image to be detected, wherein the target foreground refers to pixel points of an animal borne in the image to be detected, and the target key points comprise at least one of a mouth, an ear, a tail, front legs and rear legs of the animal borne in the image to be detected.
In some embodiments of the present application, the method further comprises:
determining a supporting surface of a target animal body according to the image to be detected, and determining a target body length of the target animal body according to the target key point, wherein the target animal body is an animal body carried in the image to be detected;
determining the projection area of the target animal body according to the supporting surface and the segmentation map;
and determining the weight of the target animal body according to the projection area and the target body length.
In some embodiments of the present application, the determining the body weight of the target animal body according to the projection area and the target body length includes:
integrating the projection area and the target body length into target one-dimensional data;
and inputting the target one-dimensional data into a preset weight measurement model so that the weight of the target animal body is regressed by the weight measurement model according to the target one-dimensional data.
In some embodiments of the present application, the weight measurement model is obtained by:
acquiring a plurality of sample data, wherein the sample data comprises the body length of a sample animal body, the projection area of the sample animal body on a supporting surface and the actual weight of the sample animal body;
integrating each sample data into one-dimensional vector data;
and training an initial model according to the one-dimensional vector data until a preset stop condition is met, and taking the trained initial model as a weight measurement model.
In some embodiments of the present application, the training an initial model according to the one-dimensional vector data includes:
inputting the one-dimensional vector data into an initial model so that the initial model regresses the predicted weight of the sample animal body according to the one-dimensional vector data;
determining a training loss value according to the predicted weight and the actual weight;
and updating the model parameters of the initial model according to the training loss value.
In some embodiments of the present application, the determining, according to the sample image, a keypoint loss value of a model to be trained includes:
acquiring a first feature extraction result of the sample image through a first network, and acquiring a feature map of the sample image through a second network;
acquiring a second feature extraction result of the feature map through the first network;
and calculating the loss between the first feature extraction result and the second feature extraction result to be used as the key point loss value of the model to be trained.
In a second aspect, an embodiment of the present application provides a training apparatus for an animal body information extraction model, where the training apparatus for an animal body information extraction model includes:
the system comprises an acquisition unit, a comparison unit and a comparison unit, wherein the acquisition unit is used for acquiring a training sample set, the training sample set comprises sample images of a plurality of animal bodies, and the sample images are marked with contour labels and a plurality of key point labels of the animal bodies;
the training unit is used for determining a segmentation loss value and a key point loss value of the model to be trained according to the sample image acquired by the acquisition unit; determining a total loss value according to the segmentation loss value and the key point loss value; and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
In some embodiments of the present application, the training apparatus for animal body information extraction model further includes a detection unit, where the detection unit is specifically configured to:
acquiring an image to be detected, and inputting the image to be detected into the animal information extraction model;
and acquiring a segmentation map and target key points of a target foreground output by the animal information extraction model according to the image to be detected, wherein the target foreground refers to pixel points of an animal borne in the image to be detected, and the target key points comprise at least one of a mouth, an ear, a tail, front legs and rear legs of the animal borne in the image to be detected.
In some embodiments of the present application, the training apparatus for animal body information extraction model further includes a processing unit, where the processing unit is specifically configured to:
determining a supporting surface of a target animal body according to the image to be detected, and determining a target body length of the target animal body according to the target key point, wherein the target animal body is an animal body carried in the image to be detected;
determining the projection area of the target animal body according to the supporting surface and the segmentation map;
and determining the weight of the target animal body according to the projection area and the target body length.
In some embodiments of the present application, the processing unit is further specifically configured to:
integrating the projection area and the target body length into target one-dimensional data;
and inputting the target one-dimensional data into a preset weight measurement model so that the weight of the target animal body is regressed by the weight measurement model according to the target one-dimensional data.
In some embodiments of the present application, the processing unit is further specifically configured to:
acquiring a plurality of sample data, wherein the sample data comprises the body length of a sample animal body, the projection area of the sample animal body on a supporting surface and the actual weight of the sample animal body;
integrating each sample data into one-dimensional vector data;
and training an initial model according to the one-dimensional vector data until a preset stop condition is met, and taking the trained initial model as a weight measurement model.
In some embodiments of the present application, the processing unit is further specifically configured to:
inputting the one-dimensional vector data into an initial model so that the initial model regresses the predicted weight of the sample animal body according to the one-dimensional vector data;
determining a training loss value according to the predicted weight and the actual weight;
and updating the model parameters of the initial model according to the training loss value.
In some embodiments of the present application, the training unit is further specifically configured to:
acquiring a first feature extraction result of the sample image through a first network, and acquiring a feature map of the sample image through a second network;
acquiring a second feature extraction result of the feature map through the first network;
and calculating the loss between the first feature extraction result and the second feature extraction result to be used as the key point loss value of the model to be trained.
In a third aspect, an embodiment of the present application further provides a training device for an animal body information extraction model, where the training device for an animal body information extraction model includes a processor and a memory, where the memory stores a computer program, and the processor executes, when calling the computer program in the memory, any one of the steps in the training method for an animal body information extraction model provided in the embodiment of the present application.
In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, where the computer program is loaded by a processor to execute the steps in the training method for the animal body information extraction model.
As can be seen from the above, the embodiments of the present application have the following beneficial effects:
the contour and the key points of the animal body borne in the sample image are labeled simultaneously, and the whole sample image, the contour and the key point labeling information are used as a training sample set of the animal body information extraction model. Because the training sample set comprises the information of the animal body in two aspects, namely the contour and the key points, the animal body information extraction model is trained through the training sample set, so that the detection of the key points is learned by the model, and meanwhile, the detection of the contour of the animal body is also learned. In view of the above two aspects of contour and key point information, model parameters are determined, so that the contour position information and key point information of the animal body are considered simultaneously in the process of detecting the key point of the animal body by the trained model, and the accuracy of detection is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic flowchart of an embodiment of a training method for an animal body information extraction model provided in an embodiment of the present application;
FIG. 2 is a schematic flow chart of detecting segmentation data and key point data of an animal body according to an embodiment of the present application;
FIG. 3 is a schematic flow chart of the method for measuring body weight of an animal according to the embodiment of the present application;
fig. 4 is a schematic flowchart of a refinement of step S90 provided in the embodiment of the present application;
FIG. 5 is a schematic structural diagram of an embodiment of a training apparatus for animal body information extraction models provided in the embodiments of the present application;
fig. 6 is a schematic structural diagram of an embodiment of a training apparatus for an animal body information extraction model provided in an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
In the description of the embodiments of the present application, it should be understood that the terms "first", "second", and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, features defined as "first", "second", may explicitly or implicitly include one or more of the described features. In the description of the embodiments of the present application, "a plurality" means two or more unless specifically defined otherwise.
The following description is presented to enable any person skilled in the art to make and use the application. In the following description, details are set forth for the purpose of explanation. It will be apparent to one of ordinary skill in the art that the present application may be practiced without these specific details. In other instances, well-known processes have not been described in detail so as not to obscure the description of the embodiments of the present application with unnecessary detail. Thus, the present application is not intended to be limited to the embodiments shown, but is to be accorded the widest scope consistent with the principles and features disclosed in the embodiments herein.
First, before describing the embodiments of the present application, the related contents of the embodiments of the present application with respect to the application context will be described.
In the technical field of computer vision, it is often necessary to identify key points of an object through an image, identify and segment a target foreground from the image, so as to further perform data processing for other purposes according to the key point information of the object and the target foreground.
For example, to measure size information of a pig body, it is often necessary to identify key points of the pig body (e.g., the pig's mouth, the pig's tail, the pig's ears, the pig's forelegs, the pig's hind legs, etc.); in order to calculate the projection area of the pig body on a certain plane, the pig body (i.e. the target foreground) needs to be identified and segmented from the image.
When a key point and a segmentation map of a pig body need to be obtained, in the prior art, target foreground segmentation and key point detection are generally performed through different models respectively. In the prior art, when the model detects the pig body key points, only the key point information of the pig body is considered (namely, the model parameters are determined only based on the key point information during model training), so that the key point detection precision is low.
Based on the above defects of the prior art, the embodiment of the present application provides a training method for an animal body information extraction model, which overcomes the defects of the prior art at least to some extent.
An execution main body of the training method for the animal information extraction model in the embodiment of the present application may be a training apparatus for the animal information extraction model provided in the embodiment of the present application, or different types of training apparatuses for the animal information extraction model, such as a server device, a physical host, or a User Equipment (UE), which are integrated with the training apparatus for the animal information extraction model, where the training apparatus for the animal information extraction model may be implemented in a hardware or software manner, and the UE may specifically be a terminal device such as a smart phone, a tablet computer, a notebook computer, a palm computer, a desktop computer, or a Personal Digital Assistant (PDA).
The training device of the animal body information extraction model can adopt a working mode of independent operation or a working mode of a device cluster, and the accuracy of the key point detection result of the key point detection model can be improved to a certain extent by applying the training method of the animal body information extraction model provided by the embodiment of the application.
In the following, a training method of an animal body information extraction model provided in an embodiment of the present application is described, in the embodiment of the present application, a training device of the animal body information extraction model is used as an execution subject, and for simplicity and convenience of description, the execution subject will be omitted in subsequent embodiments of the method, and the training method of the animal body information extraction model includes: acquiring a training sample set, wherein the training sample set comprises sample images of a plurality of animal bodies, and the sample images are marked with contour labels and a plurality of key point labels of the animal bodies; determining a segmentation loss value and a key point loss value of the model to be trained according to the sample image; determining a total loss value according to the segmentation loss value and the key point loss value; and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
Referring to fig. 1, fig. 1 is a schematic flowchart of a training method for an animal body information extraction model according to an embodiment of the present application. It should be noted that, although a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in an order different than that shown or described herein. The training method of the animal body information extraction model comprises the steps of S10-S40, wherein:
and S10, acquiring a training sample set.
The training sample set comprises a plurality of sample images, each sample image contains an animal body, and the sample images are marked with a contour label (namely contour data, which is also called segmentation data later) of the animal body and a plurality of key point labels (namely key point data). For example, the sample image is an image of a pig body, and contour data of the pig body and key point data (such as mouth, ears, tail, front legs and back legs) of the pig body are marked in the sample image. As another example, the sample image is an image of a dog, and the sample image is labeled with contour data of the dog and key point data of the dog (e.g., mouth, ears, tail, front legs, and back legs).
And S20, determining the segmentation loss value and the key point loss value of the model to be trained according to the sample image.
Mask RCNN (Regions with CNN features) is a Network architecture based on fast-Region based Convolutional Neural Network (Faster regional Convolutional Neural Network) and mainly completes example segmentation of target individuals, and Mask RCNN can be regarded as a general example segmentation architecture, which includes: classification, regression and segmentation.
The model to be trained comprises a segmentation branch and a key point detection branch, namely in the embodiment of the application, the model to be trained is obtained by improving the existing Mask RCNN network, and the specific improvement is as follows: and adding a key point detection branch on the basis of the existing Mask RCNN network structure.
The segmentation branches of the model to be trained are correspondingly provided with first loss functions, so that the model to be trained can learn the segmentation data of the animal body in the image. The first loss function is set corresponding to the split branch output split data. In the training process, the value of the first loss function is the segmentation loss value of the model to be trained.
And the key point detection branch of the model to be trained is correspondingly provided with a second loss function, so that the model to be trained can learn the key point data of the animal body in the image. The second penalty function is set corresponding to the keypoint detection branch output keypoint data. In the training process, the value of the second loss function is the loss value of the key point of the model to be trained.
The first loss function and the second loss function may be hinge loss functions (hinge loss functions), cross-entropy loss functions (cross-entropy loss functions), exponential loss functions (exponential loss functions), and the like, and specific function types of the first loss function and the second loss function are not limited in the embodiment of the present application.
Specifically, in the training process of the model to be trained, firstly, the sample image is input into the model to be trained, so that the segmentation branch of the model to be trained predicts the corresponding segmentation data according to the sample image, and the key point detection branch of the model to be trained predicts the corresponding key point data according to the sample image.
Then, a segmentation loss value is calculated through a first loss function, and the predicted segmentation data of the sample image and the segmentation data corresponding to the contour label of the sample image are substituted into the first loss function, so that the corresponding segmentation loss value can be obtained.
In some embodiments of the present application, the determining the keypoint loss value of the model to be trained from the sample images comprises the following steps a 10-a 30, wherein:
and A10, acquiring a first feature extraction result and a second feature extraction result.
The first feature extraction result refers to an RPN (Region suggestion Network) result of the sample image, and the second feature extraction result refers to an RPN result of the label image corresponding to the sample image. The RPN network has the functions of: and selecting, screening and integrating the features of the image into a feature map with a specific size.
Specifically, on one hand, the sample image is input to an RPN network of the Mask RCNN network, so that the RPN network outputs a first feature extraction result of the sample image according to the sample image. On the other hand, the label image corresponding to the sample image is input to an RPN network of the Mask RCNN network, so that the RPN network outputs a second feature extraction result of the label image according to the label image corresponding to the sample image.
And A20, calculating the loss between the first feature extraction result and the second feature extraction result to be used as the key point loss value of the model to be trained.
Specifically, firstly, representing a first feature extraction result and a second feature extraction result respectively by using the same dimension representation mode to obtain dimension representation data of the first feature extraction result and dimension representation data of the second feature extraction result; and then, calculating Softmax function loss of the first feature extraction result and the second feature extraction result according to the dimension representation data of the first feature extraction result and the dimension representation data of the second feature extraction result, and taking the Softmax function loss as a key point loss value of the model to be trained. For example, first, the RPN result of the sample image and the RPN result of the tag image corresponding to the sample image are respectively unified to the dimension of Batch _ size × Max _ detections × 7 × (56 × 56) (where Batch _ size is the Batch size and Max _ detections are the maximum number of detections). Then, Softmax function loss of the two (the RPN result of the sample image and the RPN result of the label image corresponding to the sample image) is calculated as a key point loss value of the model to be trained.
As can be derived from the above, in the embodiment of the present application, the Softmax function loss between the first feature extraction result of the sample image and the second feature extraction result of the label image corresponding to the sample image is used as the key point loss value of the model to be trained. On one hand, the first feature extraction result and the second feature extraction result are extraction data of the animal body region in the sample image and comprise the key points of the animal body, so that the accuracy of calculating the key point loss value can be ensured. On the other hand, the loss value of each key point does not need to be calculated independently, so that the calculation of the loss value of the key point can be simplified.
And S30, determining a total loss value according to the segmentation loss value and the key point loss value.
Specifically, one embodiment of step S30 is: and adding the segmentation loss value and the key point loss value to obtain a total loss value. Another embodiment of step S30 is: and further obtaining a classification loss value of the model to be trained and a regression loss value of the model to be trained, and adding the classification loss value, the regression loss value segmentation loss value and the key point loss value of the model to be trained to obtain a total loss value of the model to be trained. The classification loss value of the model to be trained is determined by the regression loss value, and the determination method of the classification loss value and the segmentation loss value of the existing Mask RCNN network can be referred to, which is not described herein again.
And S40, updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
Specifically, model parameters of the model to be trained are continuously adjusted according to the total loss value of each training until a preset training stopping condition is met, and the model to be trained at the moment is used as an animal body information extraction model. At this time, the trained animal body information extraction model can be applied to the segmentation data of the predicted animal body, and the key point data of the animal body is detected.
Wherein, the preset training stopping condition can be set according to the actual requirement. For example, it may be when the total loss value is less than a preset value; or when the total loss value is basically unchanged, namely the difference value of the total loss values corresponding to the adjacent training times is smaller than a preset value; or the iteration number of the model to be trained reaches the maximum iteration number.
From the above, in the embodiment of the present application, the contour and the key point of the animal body carried in the sample image are labeled at the same time, and the whole sample image, the contour and the key point labeling information are used as the training sample set of the animal body information extraction model. Because the training sample set comprises the information of the animal body in two aspects, namely the contour and the key points, the animal body information extraction model is trained through the training sample set, so that the detection of the key points is learned by the model, and meanwhile, the detection of the contour of the animal body is also learned. In view of the above two aspects of contour and key point information, model parameters are determined, so that the contour position information and key point information of the animal body are considered simultaneously in the process of detecting the key point of the animal body by the trained model, and the accuracy of detection is improved.
After the animal body information extraction model is obtained through the above steps S10 to S40, the animal body information extraction model may be applied to the segmentation data of the predicted animal body, and the key point data of the animal body may be detected. Referring to fig. 2, fig. 2 is a schematic flowchart of a process for detecting segmentation data and key point data of an animal body according to an embodiment of the present application. That is, in some embodiments of the present application, the training method for the animal body information extraction model further includes steps S50 to S60, where:
and S50, acquiring an image to be detected, and inputting the image to be detected into the animal information extraction model.
Wherein, the image to be detected is generally an image containing an animal body which can be identified by the animal body information extraction model
Specifically, an image to be detected is obtained, and the image to be detected is input into the animal information extraction model, so that the animal information extraction model predicts segmentation data of an animal borne in the image to be detected and key point data of the animal borne in the image to be detected according to the image to be detected.
And S60, acquiring a segmentation map of the target foreground output by the animal body information extraction model according to the image to be detected and target key points.
The target foreground refers to pixel points of an animal body borne in the image to be detected, and the target key points comprise at least one of mouth, ears, tail, front legs and rear legs of the animal body borne in the image to be detected.
After the image to be detected is input into the animal information extraction model, firstly, the classification branch of the animal information extraction model classifies the image to be recognized. After the animal body exists in the image to be detected, the animal body is segmented from the image to be detected by the segmentation branch of the animal body information extraction model, so that a segmentation map of the target foreground is obtained (namely segmentation data of the animal body in the image to be detected is obtained). And finally, detecting key point data of the animal body in the image to be detected by a key point detection branch of the animal body information extraction model according to the segmentation graph of the animal body corresponding to the image to be detected to obtain the target key point of the animal body in the image to be detected.
From the above, in the embodiment of the present application, the segmentation map and the target key points of the target foreground of the image to be detected are obtained by using the animal body information extraction model obtained through training in steps S10 to S40. The animal body information extraction model considers the information of the two aspects of the contour and the key points of the animal body at the same time, so that the segmentation graph and the target key points of the animal body in the image to be detected can be accurately and quickly detected.
After the segmented data and the key point data of the animal body are detected through the above steps S50 to S60, the segmented data and the key point data of the animal body may be applied to predict the body weight of the animal body. Referring to fig. 3, fig. 3 is a schematic flow chart of detecting body weight of an animal according to an embodiment of the present application. In some embodiments of the present application, the training method for the animal body information extraction model includes steps S70 to S90, where:
s70, determining the supporting surface of the target animal body according to the image to be detected, and determining the target body length of the target animal body according to the target key point.
Wherein, the target animal body refers to the animal body carried in the image to be detected.
"confirm the bearing surface of target animal body according to the said picture to be examined", include specifically: first, all planes existing in the image to be detected are found out by a plane fitting method (e.g., AHC algorithm of the region growing method). Then, a real plane is screened out from all planes and used as a supporting surface of the target animal body. The screening process mainly comprises two parts, namely, a plane (as a plane to be selected) with the point cloud number ratio exceeding a preset value (such as 20%) is found out from all planes; and secondly, finding out a plane with the largest depth value from all the planes to be selected, namely the real plane where the target animal body is located. Wherein, the image to be detected is a depth image.
One embodiment of "determining the target body length of the target animal body according to the target key point" is: and acquiring the distance between any two target key points as the target body length of the target animal body according to the coordinates of each point in the depth image. Wherein the target body length may be one or more.
For example, the target animal is a pig body (i.e., the animal in the image to be detected is a pig body), 7 key points (the pig's mouth, the pig's ears, the pig's tail, the pig's front legs, the pig's rear legs, and the pig's rear legs) of the pig body are detected, and the length from the pig's mouth to the pig's tail is obtained as the target body length of the pig body. And obtaining the length from the pig ear to the pig tail as the target body length of the pig body.
And S80, determining the projection area of the target animal body according to the supporting surface and the segmentation map.
Specifically, the normal vector direction of the supporting surface is taken as a projection direction, the segmentation map (i.e., the region corresponding to the target animal) is projected onto the supporting surface of the target animal, and the area of the segmentation map projected onto the supporting surface of the target animal is calculated (specifically, the calculation is performed according to the coordinates of the depth image), so as to obtain the projection area of the target animal.
And S90, determining the weight of the target animal body according to the projection area and the target body length.
Specifically, as an embodiment, the projected area of the target animal body is input into a trained weight measurement model, so that the trained weight measurement model predicts the weight of the target animal body according to the projected area of the target animal body.
As another embodiment, the target body length of the target animal body is input into the trained weight measurement model, so that the trained weight measurement model predicts the weight of the target animal body according to the target body length of the target animal body. (wherein the target body length may be one or more of, for example, the body length between the mouth and tail of the subject, and the body length between the front and rear legs of the subject, 2 body lengths together, into a trained weight measurement model)
In yet another embodiment, the target body length and the projected area of the target animal body are input into the trained weight measurement model, so that the trained weight measurement model predicts the weight of the target animal body according to the target body length and the projected area of the target animal body. (wherein the target body length may be one or more, for example, the body length between the mouth and tail of the subject, and the body length between the front and rear legs of the subject, 2 body lengths together, are input into the trained weight measurement model).
From the above, in the embodiment of the application, the target key points and the segmentation map of the target animal are detected through the animal information extraction model, the projection area of the target animal is determined based on the segmentation map, the target body length of the target animal is determined based on the segmentation map, and the weight of the target animal is determined according to the target body length and the projection area of the target animal. The weight of the animal body can be accurately and quickly measured without manually measuring the weight scale, and the carrying process of the animal body is avoided.
Referring to fig. 4, fig. 4 is a schematic flowchart of a step S90 refinement provided in the embodiment of the present application. In order to accurately detect the body weight of the target animal, in some embodiments of the present application, the step S90 specifically includes steps S91 to S92, wherein:
and S91, integrating the projection area and the target body length into target one-dimensional data.
Specifically, the target body length of the target animal body and the projection area of the target animal body on the supporting surface are described as one-dimensional vector data, and target one-dimensional data are obtained. For example, the target animal body is a pig body, and the body length of each pig body and the projection area of the pig body on the supporting surface are described as one-dimensional vector data to serve as target one-dimensional data. For another example, the sample animal is a dog, and the body length of each dog and the projection area of the dog on the supporting surface are described as one-dimensional vector data to serve as target one-dimensional data.
And S92, inputting the target one-dimensional data into a preset weight measurement model, so that the weight measurement model regresses the weight of the target animal body according to the target one-dimensional data.
The preset weight measurement model can be obtained through the following steps B10-B30, which are not described herein again. Specifically, the integrated target one-dimensional data is input into a preset weight measurement model, so that the weight measurement model predicts the weight of the target animal body according to the target body length of the target animal body and the projection area of the target animal body on the supporting surface.
As can be understood from the above, in the embodiment of the present application, the weight of the target animal body is regressed according to the target body length and the projected area of the target animal body. Due to the fact that the two aspects of the length of the target body and the projection area of the target animal body are combined at the same time, the detected weight of the target animal body is closer to the reality, and the detection precision of the weight of the target animal body is improved.
Before the weight of the target animal body is detected by using the weight measurement model in step S92, model training is required to ensure the accuracy of detecting the weight of the target animal body, in some embodiments of the present application, the weight measurement model is obtained through the following steps B10 to B30, where:
and B10, acquiring a plurality of sample data.
The sample data comprises the body length of the sample animal body, the projection area of the sample animal body on the supporting surface and the actual body weight of the sample animal body.
For convenience of understanding, the sample animal body is exemplified as a pig body. For example, each sample data includes: the length from the pig mouth to the tail is used as the body length of the sample animal body, the projection area of the pig body on the ground in the opposite direction of the normal vector, and the actual weight of the pig body (such as the weight obtained by a measuring tool).
And B20, integrating each sample data into one-dimensional vector data.
Specifically, the body length of each sample animal body, the projected area of the sample animal body on the supporting surface, and the actual body weight of the sample animal body are described as one-dimensional vector data. For example, the sample animal body is a pig body, and the body length of each pig body, the projection area of the pig body on the support surface, and the actual body weight of the pig body are described as one-dimensional vector data. In another example, the sample animal is a dog, and the body length of each dog, the projected area of the dog on the support surface, and the actual weight of the dog are described as one-dimensional vector data.
B30, training an initial model according to the one-dimensional vector data until a preset stop condition is met, and taking the trained initial model as a weight measurement model.
Specifically, the integrated one-dimensional vector data is input into an initial model, so that the initial model predicts the predicted weight of the sample animal body according to the body length of the sample animal body and the projection area of the sample animal body on a supporting surface; then, determining a training loss value of the initial model according to the predicted weight and the actual weight of the sample animal body; finally, updating the model parameters of the initial model according to the training loss value; and taking the trained initial model as a weight measurement model until the training stopping condition is met.
Wherein, the preset stop condition can be set according to the actual requirement. For example, it may be when the training loss value is less than a preset value; or when the training loss value is basically unchanged, namely the difference value of the training loss values corresponding to the adjacent training is smaller than a preset value; or the iteration number of the initial model training reaches the maximum iteration number.
From the above, in the embodiment of the application, the body length of the sample animal body, the projection area of the sample animal body on the supporting surface, and the actual weight of the sample animal body are simultaneously obtained as sample data, and model training is performed to obtain the weight measurement model. On the one hand, the body weight of the animal body can be detected by adopting the body weight measurement model only by acquiring the body length and the projection area of the animal body subsequently, the body weight scale does not need to be used manually for measurement, and the carrying process of the animal body is avoided. On the other hand, the model is trained by combining the body length and the projection area information of the animal body, so that the body weight of the animal body obtained by model learning is closer to reality, and the body weight detection precision of the model is improved.
In some embodiments of the present application, the step of training the initial model according to the one-dimensional vector data specifically includes steps B31 to B33, where:
and B31, inputting the one-dimensional vector data into an initial model, so that the initial model can regress the predicted weight of the sample animal body according to the one-dimensional vector data.
Specifically, the one-dimensional vector data is input into the initial model, so that the initial model predicts the predicted body weight of the sample animal body according to the body length of the sample animal body and the projection area of the sample animal body on the supporting surface.
And B32, determining a training loss value according to the predicted weight and the actual weight.
Specifically, the initial model is correspondingly provided with a third loss function, so that the initial model can learn the weight of the animal body from the one-dimensional vector data. The third loss function is set corresponding to the predicted body weight output by the initial model. In the training process, the value of the third loss function is the training loss value of the initial model. In the embodiment of the present application, after determining the predicted body weight and the actual body weight of the sample animal body, the training loss value of the initial model is determined according to a third loss function, which is as follows formula (1):
where L represents a training loss value of the initial model, Y represents a predicted weight of the sample animal body determined based on the one-dimensional vector data, and W represents an actual weight of the sample animal body.
And B33, updating the model parameters of the initial model according to the training loss value.
And finally, reversely propagating the training loss value of the initial model, and updating the network weight coefficient of the initial model, thereby realizing updating of the model parameters of the initial model.
In the embodiment of the present application, the network structure of the initial model is: the method is characterized by comprising three fully-connected layers and active layers which are alternately connected, wherein the number of the fully-connected layer 1, the fully-connected layer 2 and the fully-connected layer 3 can be set according to specific requirements (for example, 5, 8 and 1 in sequence), the active function adopts a ReLU function, and overfitting of a model is prevented through L2 regularization.
As can be seen from the above, in the embodiment of the present application, the training loss value is calculated by obtaining and predicting the predicted body weight and the actual body weight of the animal body according to the model; and updating the model parameters according to the training loss value, so that the prediction accuracy of the model is continuously improved, and the prediction accuracy of the trained model is further improved.
In order to better implement the training method of the animal body information extraction model in the embodiment of the present application, on the basis of the training method of the animal body information extraction model, an embodiment of the present application further provides a training device of the animal body information extraction model, as shown in fig. 5, which is a schematic structural diagram of an embodiment of the training device of the animal body information extraction model in the embodiment of the present application, and the training device 500 of the animal body information extraction model includes:
an obtaining unit 501, configured to obtain a training sample set, where the training sample set includes sample images of multiple animal bodies, and the sample images are labeled with contour labels and multiple key point labels of the animal bodies;
a training unit 502, configured to determine a segmentation loss value and a key point loss value of a model to be trained according to the sample image acquired by the acquiring unit 501; determining a total loss value according to the segmentation loss value and the key point loss value; and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
In some embodiments of the present application, the training apparatus for animal body information extraction model further includes a detection unit 503, where the detection unit 503 is specifically configured to:
acquiring an image to be detected, and inputting the image to be detected into the animal information extraction model;
and acquiring a segmentation map and target key points of a target foreground output by the animal information extraction model according to the image to be detected, wherein the target foreground refers to pixel points of an animal borne in the image to be detected, and the target key points comprise at least one of a mouth, an ear, a tail, front legs and rear legs of the animal borne in the image to be detected.
In some embodiments of the present application, the training apparatus for animal body information extraction model further includes a processing unit 504, where the processing unit 504 is specifically configured to:
determining a supporting surface of a target animal body according to the image to be detected, and determining a target body length of the target animal body according to the target key point, wherein the target animal body is an animal body carried in the image to be detected;
determining the projection area of the target animal body according to the supporting surface and the segmentation map;
and determining the weight of the target animal body according to the projection area and the target body length.
In some embodiments of the present application, the processing unit 504 is further specifically configured to:
integrating the projection area and the target body length into target one-dimensional data;
and inputting the target one-dimensional data into a preset weight measurement model so that the weight of the target animal body is regressed by the weight measurement model according to the target one-dimensional data.
In some embodiments of the present application, the processing unit 504 is further specifically configured to:
acquiring a plurality of sample data, wherein the sample data comprises the body length of a sample animal body, the projection area of the sample animal body on a supporting surface and the actual weight of the sample animal body;
integrating each sample data into one-dimensional vector data;
and training an initial model according to the one-dimensional vector data until a preset stop condition is met, and taking the trained initial model as a weight measurement model.
In some embodiments of the present application, the processing unit 504 is further specifically configured to:
inputting the one-dimensional vector data into an initial model so that the initial model regresses the predicted weight of the sample animal body according to the one-dimensional vector data;
determining a training loss value according to the predicted weight and the actual weight;
and updating the model parameters of the initial model according to the training loss value.
In some embodiments of the present application, the training unit 502 is further specifically configured to:
acquiring a first feature extraction result of the sample image through a first network, and acquiring a feature map of the sample image through a second network;
acquiring a second feature extraction result of the feature map through the first network;
and calculating the loss between the first feature extraction result and the second feature extraction result to be used as the key point loss value of the model to be trained.
In addition, in order to better implement the training method of the animal body information extraction model in the embodiment of the present application, on the basis of the training method of the animal body information extraction model, an embodiment of the present application further provides training equipment of the animal body information extraction model, referring to fig. 6, fig. 6 shows a schematic structural diagram of the training equipment of the animal body information extraction model in the embodiment of the present application, specifically, the training equipment of the animal body information extraction model provided in the embodiment of the present application includes a processor 601, and the processor 601 is configured to implement each step of the training method of the animal body information extraction model in any embodiment corresponding to fig. 1 to 4 when executing a computer program stored in a memory 602; alternatively, the processor 601 is configured to implement the functions of the units in the corresponding embodiment of fig. 5 when executing the computer program stored in the memory 602.
Illustratively, a computer program may be partitioned into one or more modules/units, which are stored in the memory 602 and executed by the processor 601 to implement embodiments of the present application. One or more modules/units may be a series of computer program instruction segments capable of performing certain functions, the instruction segments being used to describe the execution of a computer program in a computer device.
The training device for the animal body information extraction model may include, but is not limited to, a processor 601 and a memory 602. Those skilled in the art will understand that the illustration is only an example of the training device of the animal body information extraction model, and does not constitute a limitation of the training device of the animal body information extraction model, and may include more or less components than those illustrated, or combine some components, or different components, for example, the training device of the animal body information extraction model may further include an input/output device, a network access device, a bus, etc., and the processor 601, the memory 602, the input/output device, and the network access device, etc., are connected through the bus.
The Processor 601 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, discrete hardware component, etc. The general processor can be a microprocessor or the processor can be any conventional processor and the like, the processor is a control center of the training equipment of the animal body information extraction model, and various interfaces and lines are utilized to connect all parts of the training equipment of the whole animal body information extraction model.
The memory 602 may be used for storing computer programs and/or modules, and the processor 601 may implement various functions of the computer apparatus by executing or executing the computer programs and/or modules stored in the memory 602 and calling data stored in the memory 602. The memory 602 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; the storage data area may store data (such as audio data, video data, and the like) created from use of the training apparatus for the animal body information extraction model, and the like. In addition, the memory may include high speed random access memory, and may also include non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), at least one magnetic disk storage device, a Flash memory device, or other volatile solid state storage device.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the training apparatus and the device for animal body information extraction model and the corresponding units thereof described above may refer to the description of the training method for animal body information extraction model in any embodiment corresponding to fig. 1 to 4, and are not described herein again in detail.
It will be understood by those skilled in the art that all or part of the steps of the methods of the above embodiments may be performed by instructions or by associated hardware controlled by the instructions, which may be stored in a computer readable storage medium and loaded and executed by a processor.
For this reason, an embodiment of the present application provides a computer-readable storage medium, where a plurality of instructions are stored, where the instructions can be loaded by a processor to execute steps in a training method for an animal body information extraction model in any embodiment of the present application, as shown in fig. 1 to fig. 4, for specific operations, reference may be made to descriptions of the training method for an animal body information extraction model in any embodiment of fig. 1 to fig. 4, and details are not repeated here.
Wherein the computer-readable storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.
Since the instructions stored in the computer-readable storage medium can execute the steps in the method for training the animal body information extraction model in any embodiment of the present application, such as that shown in fig. 1 to 4, the beneficial effects that can be achieved by the method for training the animal body information extraction model in any embodiment of the present application, such as that shown in fig. 1 to 4, can be achieved, which are described in detail in the foregoing description and are not repeated herein.
The method, the apparatus, the device and the storage medium for training the animal body information extraction model provided in the embodiment of the present application are introduced in detail, and a specific example is applied in the present application to explain the principle and the implementation of the present application, and the description of the above embodiment is only used to help understanding the method and the core concept of the present application; meanwhile, for those skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (10)
1. A training method for animal body information extraction model is characterized in that the method comprises the following steps:
acquiring a training sample set, wherein the training sample set comprises sample images of a plurality of animal bodies, and the sample images are marked with contour labels and a plurality of key point labels of the animal bodies;
determining a segmentation loss value and a key point loss value of the model to be trained according to the sample image;
determining a total loss value according to the segmentation loss value and the key point loss value;
and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
2. The method for training animal body information extraction models according to claim 1, further comprising:
acquiring an image to be detected, and inputting the image to be detected into the animal information extraction model;
and acquiring a segmentation map and target key points of a target foreground output by the animal information extraction model according to the image to be detected, wherein the target foreground refers to pixel points of an animal borne in the image to be detected, and the target key points comprise at least one of a mouth, an ear, a tail, front legs and rear legs of the animal borne in the image to be detected.
3. The method for training animal body information extraction models according to claim 2, further comprising:
determining a supporting surface of a target animal body according to the image to be detected, and determining a target body length of the target animal body according to the target key point, wherein the target animal body is an animal body carried in the image to be detected;
determining the projection area of the target animal body according to the supporting surface and the segmentation map;
and determining the weight of the target animal body according to the projection area and the target body length.
4. The method for training animal body information extraction model according to claim 3, wherein said determining the weight of said target animal body according to said projected area and said target body length comprises:
integrating the projection area and the target body length into target one-dimensional data;
and inputting the target one-dimensional data into a preset weight measurement model so that the weight of the target animal body is regressed by the weight measurement model according to the target one-dimensional data.
5. The method for training animal body information extraction models according to claim 4, wherein the weight measurement model is obtained by:
acquiring a plurality of sample data, wherein the sample data comprises the body length of a sample animal body, the projection area of the sample animal body on a supporting surface and the actual weight of the sample animal body;
integrating each sample data into one-dimensional vector data;
and training an initial model according to the one-dimensional vector data until a preset stop condition is met, and taking the trained initial model as a weight measurement model.
6. The method for training animal body information extraction model according to claim 5, wherein said training initial model based on said one-dimensional vector data comprises:
inputting the one-dimensional vector data into an initial model so that the initial model regresses the predicted weight of the sample animal body according to the one-dimensional vector data;
determining a training loss value according to the predicted weight and the actual weight;
and updating the model parameters of the initial model according to the training loss value.
7. The method for training animal body information extraction models according to any one of claims 1-6, wherein the determining the keypoint loss value of the model to be trained according to the sample images comprises:
acquiring a first feature extraction result and a second feature extraction result, wherein the first feature extraction result refers to a region suggestion network (RPN) result of the sample image, and the second feature extraction result refers to an RPN result of a label image corresponding to the sample image;
and calculating the loss between the first feature extraction result and the second feature extraction result to be used as the key point loss value of the model to be trained.
8. A training device for animal body information extraction models is characterized by comprising:
the system comprises an acquisition unit, a comparison unit and a comparison unit, wherein the acquisition unit is used for acquiring a training sample set, the training sample set comprises sample images of a plurality of animal bodies, and the sample images are marked with contour labels and a plurality of key point labels of the animal bodies;
the training unit is used for determining a segmentation loss value and a key point loss value of the model to be trained according to the sample image acquired by the acquisition unit; determining a total loss value according to the segmentation loss value and the key point loss value; and updating the model parameters of the model to be trained according to the total loss value until a preset training stopping condition is met, and taking the trained model to be trained as an animal body information extraction model.
9. An apparatus for training animal body information extraction models, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the method for training animal body information extraction models according to any one of claims 1 to 7 when calling the computer program in the memory.
10. A computer-readable storage medium, having stored thereon a computer program which is loaded by a processor to execute the steps of the method for training an animal body information extraction model according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010211682.6A CN113449539A (en) | 2020-03-24 | 2020-03-24 | Training method, device, equipment and storage medium for animal body information extraction model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010211682.6A CN113449539A (en) | 2020-03-24 | 2020-03-24 | Training method, device, equipment and storage medium for animal body information extraction model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113449539A true CN113449539A (en) | 2021-09-28 |
Family
ID=77806394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010211682.6A Withdrawn CN113449539A (en) | 2020-03-24 | 2020-03-24 | Training method, device, equipment and storage medium for animal body information extraction model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113449539A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114550207A (en) * | 2022-01-17 | 2022-05-27 | 北京新氧科技有限公司 | Method and device for detecting key points of neck and method and device for training detection model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109657534A (en) * | 2018-10-30 | 2019-04-19 | 百度在线网络技术（北京）有限公司 | The method, apparatus and electronic equipment analyzed human body in image |
| CN109712144A (en) * | 2018-10-29 | 2019-05-03 | 百度在线网络技术（北京）有限公司 | Processing method, training method, equipment and the storage medium of face-image |
| WO2019128646A1 (en) * | 2017-12-28 | 2019-07-04 | 深圳励飞科技有限公司 | Face detection method, method and device for training parameters of convolutional neural network, and medium |
| CN110443222A (en) * | 2019-08-14 | 2019-11-12 | 北京百度网讯科技有限公司 | Method and apparatus for training face's critical point detection model |
-
2020
- 2020-03-24 CN CN202010211682.6A patent/CN113449539A/en not_active Withdrawn
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019128646A1 (en) * | 2017-12-28 | 2019-07-04 | 深圳励飞科技有限公司 | Face detection method, method and device for training parameters of convolutional neural network, and medium |
| CN109712144A (en) * | 2018-10-29 | 2019-05-03 | 百度在线网络技术（北京）有限公司 | Processing method, training method, equipment and the storage medium of face-image |
| CN109657534A (en) * | 2018-10-30 | 2019-04-19 | 百度在线网络技术（北京）有限公司 | The method, apparatus and electronic equipment analyzed human body in image |
| CN110443222A (en) * | 2019-08-14 | 2019-11-12 | 北京百度网讯科技有限公司 | Method and apparatus for training face's critical point detection model |
Non-Patent Citations (2)
| Title |
|---|
| 张凯中;朱伟兴;: "基于改进Mask RCNN的俯视群养猪图像的分割", 软件, no. 03, 15 March 2020 (2020-03-15) * |
| 高云;郭继亮;黎煊;雷明刚;卢军;童宇;: "基于深度学习的群猪图像实例分割方法", 农业机械学报, no. 04, 5 March 2019 (2019-03-05) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114550207A (en) * | 2022-01-17 | 2022-05-27 | 北京新氧科技有限公司 | Method and device for detecting key points of neck and method and device for training detection model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11557085B2 (en) | Neural network processing for multi-object 3D modeling | |
| CN108038474B (en) | Face detection method, convolutional neural network parameter training method, device and medium | |
| US20200356818A1 (en) | Logo detection | |
| US20200074665A1 (en) | Object detection method, device, apparatus and computer-readable storage medium | |
| CN112330601A (en) | Parking detection method, device, equipment and medium based on fisheye camera | |
| JP2022532460A (en) | Model training methods, equipment, terminals and programs | |
| CN111985458B (en) | Method for detecting multiple targets, electronic equipment and storage medium | |
| CN111626295B (en) | Training method and device for license plate detection model | |
| US11972578B2 (en) | Method and system for object tracking using online training | |
| CN113449538A (en) | Visual model training method, device, equipment and storage medium | |
| CN111444807A (en) | Target detection method, device, electronic equipment and computer readable medium | |
| CN107993239B (en) | Method and device for calculating depth order of monocular image | |
| CN110956131A (en) | Single-target tracking method, device and system | |
| WO2022127333A1 (en) | Training method and apparatus for image segmentation model, image segmentation method and apparatus, and device | |
| CN113449539A (en) | Training method, device, equipment and storage medium for animal body information extraction model | |
| CN114187465A (en) | Method and device for training classification model, electronic equipment and storage medium | |
| CN110910478B (en) | GIF map generation method and device, electronic equipment and storage medium | |
| CN117058421A (en) | Multi-head model-based image detection key point method, system, platform and medium | |
| US20230401670A1 (en) | Multi-scale autoencoder generation method, electronic device and readable storage medium | |
| CN116309643A (en) | Face shielding score determining method, electronic equipment and medium | |
| CN114998743A (en) | Method, device, equipment and medium for constructing visual map points | |
| CN115004245A (en) | Target detection method, target detection device, electronic equipment and computer storage medium | |
| CN113705643A (en) | Target detection method and device and electronic equipment | |
| TWI840833B (en) | Method for generating multiscale autoencoder, computer device and storage medium | |
| US20240107986A1 (en) | Fish identification device and fish identification method |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20210928 |