CN114445684A - Method, device and equipment for training lane line segmentation model and storage medium - Google Patents
Method, device and equipment for training lane line segmentation model and storage medium Download PDFInfo
- Publication number
- CN114445684A CN114445684A CN202210183501.2A CN202210183501A CN114445684A CN 114445684 A CN114445684 A CN 114445684A CN 202210183501 A CN202210183501 A CN 202210183501A CN 114445684 A CN114445684 A CN 114445684A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- loss function
- training
- feature
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method, a device, equipment and a storage medium for training a lane line segmentation model, relates to the field of machine learning, and solves the problem of more consumption of calculation resources during lane line position detection. The training method of the lane line segmentation model comprises the following steps: acquiring a training picture of a lane line; respectively inputting the training pictures into a first model and a second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model; obtaining an imitation loss function according to the first characteristic data and the second characteristic data; obtaining a classification loss function according to the label data and the second characteristic data of the training picture; the second model is trained according to the modeling loss function and the classification loss function.
Description
Technical Field
The invention relates to the field of machine learning, in particular to a method, a device, equipment and a storage medium for training a lane line segmentation model.
Background
With the development of science and technology, the automatic driving technology gradually advances into the daily life of people, and the travel mode of people is changed unconsciously. The accuracy of lane line position detection in the automatic driving technique affects the accuracy of path planning.
In the prior art, in order to reduce resource consumption in the detection of the position of the lane line, a lightweight detection model is generally adopted to detect the position of the lane line, but the lightweight detection model is generally limited in learning capacity, so that the problem that the distribution of positive and negative samples in the detection of the lane line is seriously unbalanced can not be solved well, and the problem of low detection precision exists.
Disclosure of Invention
The application provides a method, a device, equipment and a storage medium for training a lane line segmentation model, which can improve the detection precision of a lightweight lane line position detection model.
In a first aspect of the embodiments of the present application, a method for training a lane line segmentation model is provided, where the method includes: acquiring a training picture of a lane line; respectively inputting the training pictures into a first model and a second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model; obtaining a simulation loss function according to the first characteristic data and the second characteristic data; obtaining a classification loss function according to the training picture label data and the second characteristic data of the lane line; the second model is trained according to the modeling loss function and the classification loss function.
In one embodiment, deriving the emulated loss function from the first characterization data and the second characterization data comprises:
respectively carrying out feature fusion processing on the first feature data and the second feature data to obtain first fusion data and second fusion data;
and obtaining the simulation loss function according to the first fusion data and the second fusion data.
In one embodiment, deriving the emulated loss function from the first fused data and the second fused data comprises:
performing difference operation on the first fusion data and the second fusion data to obtain characteristic difference;
respectively inputting the first fusion data and the second fusion data into a preset classification network to obtain first target data and second target data;
and obtaining the simulation loss function according to the characteristic difference, the first target data and the second target data.
In one embodiment, deriving the emulated loss function from the feature difference, the first target data, and the second target data comprises:
respectively carrying out maximum value operation on the first target data and the second target data to obtain a first mask and a second mask;
a simulated loss function is derived from the feature difference, the first mask, and the second mask.
In one embodiment, deriving the simulated loss function from the feature difference, the first mask, and the second mask comprises:
performing OR operation on the first mask and the second mask to obtain a simulated mask;
a simulated loss function is derived from the feature differences and the simulated mask.
In one embodiment, obtaining the classification loss function according to the training picture label data and the second feature data of the lane line includes: performing feature fusion processing on the second feature data to obtain second fusion data; respectively inputting the second fusion data into a preset classification model to obtain a model classification result; acquiring label data of a lane line training picture; and obtaining a classification loss function according to the model classification result and the label data.
In one embodiment, the second model includes a full convolution adaptation layer, the method further comprising:
and utilizing a second model comprising a full convolution adaptation layer to extract the features of the training picture.
In a second aspect of the embodiments of the present application, a training device for a lane line segmentation model is provided, where the training device includes:
the acquisition module is used for acquiring a training picture of the lane line;
the extraction module is used for inputting the training pictures into the first model and the second model respectively to carry out feature extraction so as to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model;
the first processing module is used for obtaining a simulation loss function according to the first characteristic data and the second characteristic data;
the second processing module is used for obtaining a classification loss function according to the training picture label data and the second characteristic data of the lane line;
and the training module is used for training the second model according to the simulation loss function and the classification loss function.
In a third aspect of the embodiments of the present application, a computer device is provided, where the computer device includes a memory and a processor, and the memory stores a computer program, and when the computer program is executed by the processor, the computer program implements the method for training a lane line segmentation model according to the first aspect of the embodiments of the present application.
In a fourth aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program, when executed by a processor, implements the method for training a lane line segmentation model according to the first aspect of the embodiments of the present application.
The training method of the lane line segmentation model provided by the embodiment of the application obtains the training picture of the lane line, respectively inputting the training pictures into the first model and the second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model, then the simulation loss function is obtained according to the first characteristic data and the second characteristic data, and obtaining a classification loss function according to the training picture label data and the second characteristic data of the lane line, and finally obtaining a classification loss function according to the simulation loss function and the classification loss function, the second model is trained, so that the second model is trained through the simulation loss function and the classification loss function, the second model simulates the detection capability of the first model, the detection precision of the light-weight lane line position detection model can be improved, and meanwhile, the consumption of computing resources can be reduced.
Drawings
Fig. 1 is a schematic internal structural diagram of a computer device according to an embodiment of the present application;
fig. 2 is a flowchart of a method for training a lane line segmentation model according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a simulated mask generation process provided by an embodiment of the present application;
fig. 4 is a first schematic diagram illustrating training of a lane line segmentation model according to an embodiment of the present disclosure;
fig. 5 is a second schematic diagram illustrating training of a lane line segmentation model according to an embodiment of the present disclosure;
fig. 6 is a structural diagram of a training device for a lane line segmentation model according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, 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 invention.
In the following, the terms "first", "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of the present disclosure, "a plurality" means two or more unless otherwise specified.
In addition, the use of "based on" or "according to" means open and inclusive, as a process, step, calculation, or other action that is "based on" or "according to" one or more conditions or values may in practice be based on additional conditions or values beyond those that are present.
In order to solve the problem of high consumption of computing resources during lane line position detection, the embodiment of the application provides a training method of a lane line segmentation model, which obtains first feature data and second feature data by obtaining a training picture of a lane line and inputting the training picture into a first model and a second model respectively for feature extraction, wherein the number of model layers of the first model is larger than that of the second model, then obtains a simulation loss function according to the first feature data and the second feature data, obtains a classification loss function according to training picture label data and the second feature data of the lane line, and finally trains the second model according to the simulation loss function and the classification loss function, so that the second model is trained through the simulation loss function and the classification loss function to enable the second model to simulate the detection capability of the first model, and the detection precision of a light-weight lane line position detection model can be improved, meanwhile, the consumption of computing resources can be reduced.
The execution main body of the training method for the lane line segmentation model provided by the embodiment of the application can be computer equipment, terminal equipment or a server, wherein the terminal equipment can be a vehicle-mounted terminal, various personal computers, notebook computers, smart phones, tablet computers, portable wearable equipment and the like, and the comparison of the application is not particularly limited.
Fig. 1 is a schematic internal structural diagram of a computer device according to an embodiment of the present disclosure. As shown in fig. 1, the computer device includes a processor and a memory connected by a system bus. Wherein the processor is configured to provide computational and control capabilities. The memory may include a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The computer program can be executed by a processor for implementing the steps of a training method of a lane line segmentation model provided in the above embodiments. The internal memory provides a cached execution environment for the operating system and computer programs in the non-volatile storage medium.
Those skilled in the art will appreciate that the architecture shown in fig. 1 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
Based on the execution main body, the embodiment of the application provides a training method of a lane line segmentation model. As shown in fig. 2, the method comprises the steps of:
Wherein, the training picture can be the actual road picture of gathering.
And the model layer number of the first model is larger than that of the second model. The first model may also be referred to as a teacher model and the second model may also be referred to as a student model.
Optionally, the first model may be a network model of ResNet101, and the second model may be a network model of ResNet18, and the number of layers of the specific network model is not limited in the present application.
And step 203, obtaining a simulation loss function according to the first characteristic data and the second characteristic data.
And step 204, obtaining a classification loss function according to the label data and the second characteristic data of the training picture of the lane line.
The label data is used for indicating the real classification result of the lane line training picture.
The training method of the lane line segmentation model provided by the embodiment of the application obtains the training picture of the lane line, respectively inputting the training pictures into the first model and the second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model, then the simulation loss function is obtained according to the first characteristic data and the second characteristic data, and obtaining a classification loss function according to the training picture label data and the second characteristic data of the lane line, and finally obtaining a classification loss function according to the simulation loss function and the classification loss function, the second model is trained, so that the second model is trained through the simulation loss function and the classification loss function, the second model simulates the detection capability of the first model, the detection precision of the light-weight lane line position detection model can be improved, and meanwhile, the consumption of computing resources can be reduced.
In one embodiment, deriving the emulated loss function from the first characterization data and the second characterization data comprises: respectively carrying out feature fusion processing on the first feature data and the second feature data to obtain first fusion data and second fusion data; and obtaining the simulation loss function according to the first fusion data and the second fusion data.
Optionally, after the training pictures are respectively subjected to feature extraction to obtain first feature data and second feature data, the first feature data and the second feature data may be input to a preset feature fusion network to be subjected to feature fusion, so as to respectively obtain first fusion data and second fusion data.
In one embodiment, deriving the emulated loss function from the first fused data and the second fused data comprises:
and performing difference operation on the first fusion data and the second fusion data to obtain a characteristic difference, inputting the first fusion data and the second fusion data to a preset classification network respectively to obtain first target data and second target data, and obtaining a simulation loss function according to the characteristic difference, the first target data and the second target data.
Optionally, obtaining the simulated loss function according to the feature difference, the first target data, and the second target data includes: respectively carrying out maximum value operation on the first target data and the second target data to obtain a first mask and a second mask; a simulated loss function is derived from the feature difference, the first mask, and the second mask.
In one embodiment, deriving the simulated loss function from the feature difference, the first mask, and the second mask comprises: performing OR operation on the first mask and the second mask to obtain a simulated mask; a simulated loss function is derived from the feature differences and the simulated mask.
In a specific implementation process, a prediction graph is obtained after a first target graph formed by first target data output by a teacher model is subjected to a large value calculation, and a threshold value F is calculated for Mt:
based on the threshold F, those positions with predicted values higher than F can be filtered out, resulting in a first mask of W × H.
After the output of the student model is subjected to large value calculation, a prediction graph is also obtained and is counted as Ms, and a threshold value F is calculated:
based on the threshold F, those positions with predicted values higher than F can be filtered out, resulting in a second mask of W × H. Then, the first mask and the second mask are ORed to obtain a final fine-grained estimated simulated mask. The simulated mask is generated as shown in fig. 3. The numbers in the pictures represent pixel values in the prediction graph, after maximum value operation, the prediction graph is reserved with the larger corresponding pixel value to obtain the prediction graph, the prediction graph is taken to be greater than a threshold value 6 and is 1, and less than or equal to 6 and is 0 to obtain a mask graph, similarly, the mask graph of the student model can be obtained, and then OR operation is carried out on 4 and 8 to obtain a final fine-grained simulated mask graph.
Defining s as a student model characteristic diagram and t as a teacher model characteristic diagram, and defining an objective function for each lane line adjacent region (i, j) on the characteristic diagram, and training the student model to minimize the objective function as follows.
Where (i, j) represents the location, c represents the channel, and fadap represents the full convolution adaptation layer.
The ability to learn the teacher model to extract features minimizes the distance of the student network from the teacher network's feature response map at all estimated locations, i.e., the final simulated loss function, as follows:
wherein W, H are width and height of the feature map, Np is normalized value, IijIs mask, other parameters are as above. The total loss that ultimately serves to train the student network: the weight parameter lambda adopts a self-adaptive parameter, and when the simulation loss function value is large, the value is large; when the simulated loss function value is small, a small value is taken.
L＝Lgt+λLimitation
Where Lgt is the loss function of the lane line classification and the simulated loss function is the loss function of the knowledge distillation. The second network is trained by classifying a loss function and modeling the loss function.
In one embodiment, obtaining the classification loss function according to the training picture label data and the second feature data of the lane line includes: performing feature fusion processing on the second feature data to obtain second fusion data; inputting the second fusion data into a preset classification network respectively to obtain a model classification result; acquiring label data of a lane line training picture; and obtaining a classification loss function according to the model classification result and the label data.
Optionally, as shown in fig. 4, before the first fused data and the second fused data are respectively input to the preset classification network, the first fused data and the second fused data may be respectively encoded, and then the encoded first fused data and the encoded second fused data are respectively input to the preset classification network.
In one embodiment, the second model includes a full convolution adaptation layer, the method further comprising: and utilizing a second model comprising a full convolution adaptation layer to extract the features of the training picture.
As shown in fig. 5, by adding a full convolution adaptation layer behind the student network, the number of channels of the feature response graphs of the teacher network and the student network can be unified, the performance can be improved, and feature extraction can be facilitated.
In order to facilitate understanding of those skilled in the art, an embodiment of the present application further provides a method for training a lane line segmentation model, and specifically, the method includes:
(1) acquiring a training picture of a lane line;
(2) respectively inputting the training pictures into a first model and a second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model;
(3) respectively carrying out feature fusion processing on the first feature data and the second feature data to obtain first fusion data and second fusion data;
(4) performing difference operation on the first fusion data and the second fusion data to obtain characteristic difference;
(5) respectively inputting the first fusion data and the second fusion data into a preset classification network to obtain first target data and second target data;
(6) respectively carrying out maximum value operation on the first target data and the second target data to obtain a first mask and a second mask;
(7) performing OR operation on the first mask and the second mask to obtain a simulated mask;
(8) a simulated loss function is derived from the feature differences and the simulated mask.
(9) Performing feature fusion processing on the second feature data to obtain second fusion data;
(10) inputting the second fusion data into a preset classification model respectively to obtain classification results;
(11) acquiring label data of a lane line training picture;
(12) and obtaining a classification loss function according to the model classification result and the label data.
(13) The second model is trained according to the modeling loss function and the classification loss function.
The training method of the lane line segmentation model provided by the embodiment of the application obtains the training picture of the lane line, respectively inputting the training pictures into the first model and the second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model, then the simulation loss function is obtained according to the first characteristic data and the second characteristic data, and obtaining a classification loss function according to the label data and the second characteristic data of the lane line, and finally obtaining a classification loss function according to the simulation loss function and the classification loss function, the second model is trained, so that the second model is trained through the simulation loss function and the classification loss function, the second model simulates the detection capability of the first model, the detection precision of the light-weight lane line position detection model can be improved, and meanwhile, the consumption of computing resources can be reduced.
It should be noted that the training method of the lane line segmentation model provided in the embodiment of the present application can be applied in a complementary manner by being superimposed on schemes such as distillation, self-attention distillation, and temperature detection of new distillation, and the situation that the new distillation cannot be applied in a superimposed manner is avoided.
Further, in order to solve the problem of imbalance between positive and negative samples in lane line detection, the method for training a lane line segmentation model provided by the application takes a teacher model and a lane line positive sample predicted in a student network as a reference, and takes the gradient continuity of an image into consideration, and a region in a range close to the lane line positive sample is a pixel region which is difficult to classify. The lane lines and the nearby areas form masks to extract the difference between the feature graph of each channel of the teacher model and the corresponding feature graph of the student model, and the student model is guided to mainly learn the capability of the teacher model for extracting the features of the lane lines and the feature extracting capability of the nearby areas of the lane lines, but not the capability of learning all the channels of the teacher trunk feature extracting network for extracting the features of the whole feature graph, so that the noise introduced by the background area is reduced, and the detection accuracy is improved.
For the implementation processes of (1) to (13), reference may be specifically made to the description of the above embodiments, and the implementation principles and technical effects thereof are similar and are not described herein again. It should be understood that the steps in the step flow chart in the embodiment of the training method of each lane line segmentation model described above are sequentially displayed as indicated by the arrow, but the steps are not necessarily sequentially executed in the order indicated by the arrow. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least a part of the steps of the above-mentioned flowcharts may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, and the order of performing the steps or stages is not necessarily sequential, but may be performed alternately or alternately with other steps or at least a part of the steps or stages in other steps.
The scheme provided by the embodiment of the application is mainly introduced from the perspective of a method. To implement the above functions, it includes hardware structures and/or software modules for performing the respective functions. Those of skill in the art would readily appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as hardware or combinations of hardware and computer software. Whether a function is performed as hardware or computer software drives hardware depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
As shown in fig. 6, an embodiment of the present application provides a training apparatus for a lane line segmentation model, including: an acquisition module 11, an extraction module 12, a first processing module 13, a second processing module 14 and a training module 15.
The acquisition module 11 is used for acquiring a training picture of a lane line;
the extraction module 12 is configured to input the training pictures into the first model and the second model respectively to perform feature extraction, so as to obtain first feature data and second feature data, where the number of model layers of the first model is greater than that of the second model;
the first processing module 13 is configured to obtain a simulated loss function according to the first characteristic data and the second characteristic data;
the second processing module 14 is configured to obtain a classification loss function according to the training image label data and the second feature data of the lane line;
a training module 15, configured to train the second model according to the modeling loss function and the classification loss function.
In one embodiment, the first processing module 13 is specifically configured to:
respectively carrying out feature fusion processing on the first feature data and the second feature data to obtain first fusion data and second fusion data;
and obtaining the simulation loss function according to the first fusion data and the second fusion data.
In one embodiment, the first processing module 13 is specifically configured to:
performing difference operation on the first fusion data and the second fusion data to obtain characteristic difference;
respectively inputting the first fusion data and the second fusion data into a preset classification network to obtain first target data and second target data;
and obtaining the simulation loss function according to the characteristic difference, the first target data and the second target data.
In an embodiment, the first processing module 13 is specifically configured to:
respectively carrying out maximum value operation on the first target data and the second target data to obtain a first mask and a second mask;
a simulated loss function is derived from the feature difference, the first mask, and the second mask.
In one embodiment, the first processing module 13 is specifically configured to:
performing OR operation on the first mask and the second mask to obtain a simulated mask;
a simulated loss function is derived from the feature differences and the simulated mask.
In one embodiment, the second processing module 14 is specifically configured to:
performing feature fusion processing on the second feature data of the feature difference to obtain second fusion data;
respectively inputting the second fusion data into a preset classification model to obtain a model classification result;
acquiring label data of a lane line training picture;
and obtaining a classification loss function according to the model classification result and the label data.
In one embodiment, the second model comprises a full convolution adaptation layer, and the extraction module 12 is further configured to: and performing feature extraction on the training picture by using a second model comprising a full convolution adaptation layer.
The training device for the lane line segmentation model, provided by the embodiment of the application, obtains the first characteristic data and the second characteristic data by obtaining the training pictures of the lane lines and inputting the training pictures into the first model and the second model respectively for characteristic extraction, wherein the number of model layers of the first model is larger than that of the second model, then obtains the simulation loss function according to the first characteristic data and the second characteristic data, obtains the classification loss function according to the training picture label data and the second characteristic data of the lane lines, and finally trains the second model according to the simulation loss function and the classification loss function, so that the second model is trained through the simulation loss function and the classification loss function, the detection capability of the second model is removed from the simulation loss function, the detection precision of the light-weight lane line position detection model can be improved, and simultaneously, the consumption of computing resources can also be reduced.
The training device for the lane line segmentation model provided by this embodiment may implement the above method embodiments, and its implementation principle and technical effect are similar, which are not described herein again.
For specific limitations of the training device for the lane line segmentation model, reference may be made to the above limitations of the training method for the lane line segmentation model, and details are not repeated here. All or part of each module in the training device of the lane line segmentation model can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent of a processor in the server, and can also be stored in a memory in the server in a software form, so that the processor can call and execute operations corresponding to the modules.
In another embodiment of the present application, a computer device is further provided, which includes a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to implement the steps of the training method of the lane line segmentation model according to the embodiment of the present application.
In another embodiment of the present application, a computer-readable storage medium is further provided, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the training method of the lane line segmentation model according to the embodiment of the present application.
In another embodiment of the present application, a computer program product is further provided, where the computer program product includes computer instructions that, when run on a training apparatus for a lane line segmentation model, cause the training apparatus for the lane line segmentation model to perform the steps performed by the method for training a lane line segmentation model in the method flow shown in the above-mentioned method embodiment.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented using a software program, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. The processes or functions according to the embodiments of the present application are generated in whole or in part when the computer-executable instructions are loaded and executed on a computer. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored on a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). Computer-readable storage media can be any available media that can be accessed by a computer or can comprise one or more data storage devices, such as servers, data centers, and the like, that can be integrated with the media. The usable medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., Solid State Disk (SSD)), among others.
All possible combinations of the technical features in the above embodiments may not be described for the sake of brevity, but should be considered as being within the scope of the present disclosure as long as there is no contradiction between the combinations of the technical features.
The above examples only show some embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A method for training a lane line segmentation model, the method comprising:
acquiring a training picture of a lane line;
respectively inputting the training pictures into a first model and a second model for feature extraction to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model;
obtaining a simulation loss function according to the first characteristic data and the second characteristic data;
obtaining a classification loss function according to the label data of the training picture of the lane line and the second characteristic data;
and training the second model according to the simulation loss function and the classification loss function.
2. The method of claim 1, wherein deriving an emulated loss function from the first and second characterization data comprises:
respectively performing feature fusion processing on the first feature data and the second feature data to obtain first fusion data and second fusion data;
and obtaining the simulated loss function according to the first fusion data and the second fusion data.
3. The method of claim 2, wherein said deriving the mimic loss function from the first fused data and the second fused data comprises:
performing difference operation on the first fusion data and the second fusion data to obtain a characteristic difference;
inputting the first fusion data and the second fusion data into a preset classification network respectively to obtain first target data and second target data;
and obtaining the simulation loss function according to the characteristic difference, the first target data and the second target data.
4. The method of claim 3, wherein deriving the mimic loss function from the feature difference, the first target data, and the second target data comprises:
respectively carrying out maximum value operation on the first target data and the second target data to obtain a first mask and a second mask;
obtaining the mimic loss function from the feature difference, the first mask, and the second mask.
5. The method of claim 4, wherein deriving the mimic loss function from the feature differences, the first mask, and the second mask comprises:
performing an OR operation on the first mask and the second mask to obtain a simulated mask;
the mimic loss function is derived from the feature differences and the mimic mask.
6. The method of claim 1, wherein obtaining a classification loss function according to the label data and the second feature data of the training picture of the lane marking comprises:
performing feature fusion processing on the second feature data to obtain second fusion data;
inputting the second fusion data into preset classification models respectively to obtain model classification results;
acquiring label data of the lane line training picture;
and obtaining the classification loss function according to the model classification result and the label data.
7. The method of claim 1, wherein the second model comprises a full convolution adaptation layer, the method further comprising:
and utilizing a second model comprising the full convolution adaptation layer to perform feature extraction on the training picture.
8. A training apparatus for a lane line segmentation model, the apparatus comprising:
the acquisition module is used for acquiring a training picture of the lane line;
the extraction module is used for inputting the training pictures into a first model and a second model respectively to carry out feature extraction so as to obtain first feature data and second feature data, wherein the number of model layers of the first model is larger than that of the second model;
the first processing module is used for obtaining a simulation loss function according to the first characteristic data and the second characteristic data;
the second processing module is used for obtaining a classification loss function according to the training picture label data of the lane line and the second characteristic data;
and the training module is used for training the second model according to the simulation loss function and the classification loss function.
9. A computer arrangement comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, implements the method of training a lane marking segmentation model according to any one of claims 1 to 7.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, implements the method of training a lane line segmentation model according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210183501.2A CN114445684A (en) | 2022-02-25 | 2022-02-25 | Method, device and equipment for training lane line segmentation model and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210183501.2A CN114445684A (en) | 2022-02-25 | 2022-02-25 | Method, device and equipment for training lane line segmentation model and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114445684A true CN114445684A (en) | 2022-05-06 |
Family
ID=81374363
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210183501.2A Pending CN114445684A (en) | 2022-02-25 | 2022-02-25 | Method, device and equipment for training lane line segmentation model and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114445684A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115393595A (en) * | 2022-10-27 | 2022-11-25 | 福思(杭州)智能科技有限公司 | Segmentation network model training method, lane line detection method and electronic device |
| CN115565148A (en) * | 2022-11-09 | 2023-01-03 | 福思(杭州)智能科技有限公司 | Road image detection method, road image detection device, storage medium and electronic device |
-
2022
- 2022-02-25 CN CN202210183501.2A patent/CN114445684A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115393595A (en) * | 2022-10-27 | 2022-11-25 | 福思(杭州)智能科技有限公司 | Segmentation network model training method, lane line detection method and electronic device |
| CN115393595B (en) * | 2022-10-27 | 2023-02-03 | 福思(杭州)智能科技有限公司 | Segmentation network model training method, lane line detection method, device and medium |
| CN115565148A (en) * | 2022-11-09 | 2023-01-03 | 福思(杭州)智能科技有限公司 | Road image detection method, road image detection device, storage medium and electronic device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | An edge traffic flow detection scheme based on deep learning in an intelligent transportation system | |
| WO2021232985A1 (en) | Facial recognition method and apparatus, computer device, and storage medium | |
| CN109658418A (en) | Learning method, device and the electronic equipment of scene structure | |
| CN114445684A (en) | Method, device and equipment for training lane line segmentation model and storage medium | |
| CN114912612A (en) | Bird identification method and device, computer equipment and storage medium | |
| CN113177559B (en) | Image recognition method, system, equipment and medium combining breadth and dense convolutional neural network | |
| CN107247952B (en) | Deep supervision-based visual saliency detection method for cyclic convolution neural network | |
| CN113792871A (en) | Neural network training method, target identification method, device and electronic equipment | |
| CN109583367A (en) | Image text row detection method and device, storage medium and electronic equipment | |
| CN115082752A (en) | Target detection model training method, device, equipment and medium based on weak supervision | |
| CN110807379A (en) | Semantic recognition method and device and computer storage medium | |
| CN115953643A (en) | Knowledge distillation-based model training method and device and electronic equipment | |
| CN111008622B (en) | Image object detection method and device and computer readable storage medium | |
| CN116701706A (en) | Data processing method, device, equipment and medium based on artificial intelligence | |
| CN110263842A (en) | For the neural network training method of target detection, device, equipment, medium | |
| CN116958729A (en) | Training of object classification model, object classification method, device and storage medium | |
| CN113743448B (en) | Model training data acquisition method, model training method and device | |
| CN112862840B (en) | Image segmentation method, device, equipment and medium | |
| CN117036843A (en) | Target detection model training method, target detection method and device | |
| CN117010480A (en) | Model training method, device, equipment, storage medium and program product | |
| CN116546304A (en) | Parameter configuration method, device, equipment, storage medium and product | |
| CN113824989A (en) | Video processing method and device and computer readable storage medium | |
| CN111091198A (en) | Data processing method and device | |
| CN117542085B (en) | Park scene pedestrian detection method, device and equipment based on knowledge distillation | |
| CN113869516B (en) | Knowledge graph embedded model training method and device, electronic equipment and 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 |