CN113361511A - Method, device and equipment for establishing correction model and computer readable storage medium - Google Patents
Method, device and equipment for establishing correction model and computer readable storage medium Download PDFInfo
- Publication number
- CN113361511A CN113361511A CN202010147659.5A CN202010147659A CN113361511A CN 113361511 A CN113361511 A CN 113361511A CN 202010147659 A CN202010147659 A CN 202010147659A CN 113361511 A CN113361511 A CN 113361511A
- Authority
- CN
- China
- Prior art keywords
- category identification
- commodity category
- commodity
- identification result
- model
- 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
- 238000012937 correction Methods 0.000 title claims abstract description 92
- 238000000034 method Methods 0.000 title claims abstract description 64
- 238000012549 training Methods 0.000 claims abstract description 61
- 238000005381 potential energy Methods 0.000 claims description 54
- 230000006870 function Effects 0.000 claims description 40
- 238000004590 computer program Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 238000010586 diagram Methods 0.000 description 10
- 238000013473 artificial intelligence Methods 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 6
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000003550 marker Substances 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000000137 annealing Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000002922 simulated annealing Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000013068 supply chain management Methods 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
-
- 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)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (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 provides a method, a device and equipment for establishing a correction model and a computer readable storage medium, wherein the generated correction model can correct error identification results in commodity type identification results given by a commodity type identification model. The method provided by the embodiment of the application comprises the following steps: acquiring a commodity category identification result set, wherein the commodity category identification result set comprises commodity category identification results of commodities contained in different images to be identified, and the commodity category identification results comprise marked wrong commodity category identification results and commodity category identification results corrected by the wrong commodity category identification results; sequentially carrying out grid distribution on the commodity category identification result of the commodity contained in each image to be identified according to the placement position of the commodity to obtain a plurality of grid images; and performing model training on the initial model through a plurality of grid images by taking the error commodity category identification result in the input grid image as a target, and taking the trained model as a correction model.
Description
Technical Field
The present application relates to the field of image recognition technologies, and in particular, to a method, an apparatus, a device, and a computer-readable storage medium for building a correction model.
Background
In recent years, Artificial Intelligence (AI) has seen explosive growth in various fields, and from the technical field, the application of AI technology has focused more on voice and vision, such as intelligent driving, unmanned aerial vehicle, Augmented Reality (AR), Virtual Reality (VR), big data and data services. Meanwhile, the AI technology has the problem that the accuracy is still to be improved, is still in the weak AI stage in partial application fields, and has a long way to be moved away from the comprehensive integration of the technical field and the large-scale popularization and application.
Taking a commodity category identification scene in image identification as an example, in the existing related technology, a large number of commodity images can be acquired to train a commodity category identification model based on an AI technology, however, when the trained model is put into practical application, the problem of mistakenly identifying commodities still exists, especially the situations that the actual scene is disordered, the similarity of adjacent commodities is high, and the commodities are hidden or reflected, and the identification processing of the commodity category identification model is still a huge challenge.
Therefore, if the accuracy of identifying the product type can be improved, the AI technique can play a more powerful role in product type identification, supermarket management, and even supply chain management.
Disclosure of Invention
The embodiment of the application provides a method, a device and equipment for establishing a correction model and a computer readable storage medium, wherein the generated correction model can correct an error identification result in a commodity type identification result given by a commodity type identification model, so that the identification precision of the commodity type identification result is improved.
In a first aspect, an embodiment of the present application provides a method for building a correction model, where the method includes:
acquiring a commodity category identification result set, wherein the commodity category identification result set comprises commodity category identification results of commodities contained in different images to be identified, and the commodity category identification results comprise marked wrong commodity category identification results and commodity category identification results corrected by the wrong commodity category identification results;
sequentially carrying out grid distribution on the commodity category identification result of the commodity contained in each image to be identified according to the placement position of the commodity to obtain a plurality of grid images;
and performing model training on the initial model through a plurality of grid images by taking the error commodity category identification result in the input grid image as a target, and taking the trained model as a correction model.
With reference to the first aspect of the embodiment of the present application, in a first possible implementation manner of the first aspect of the embodiment of the present application, sequentially performing grid distribution on the product category identification result of the product included in each image to be identified according to the placement position of the product, and obtaining a plurality of grid images includes:
sequentially extracting commodity type labels from the commodity type identification result of commodities contained in each image to be identified;
and carrying out grid distribution on the multiple groups of commodity category labels according to the placing positions of the commodities to obtain multiple grid images.
With reference to the first aspect of the embodiment of the present application, in a second possible implementation manner of the first aspect of the embodiment of the present application, performing model training on the initial model through a plurality of mesh images includes:
calculating a first potential energy through a Markov random field based on the corrected commodity category identification result of the false commodity category identification result in each grid image, wherein the first potential energy is used for indicating the degree of difference between different commodity category identification results;
and carrying out model training on the initial model through the plurality of grid images and the first loss function.
With reference to the second possible implementation manner of the first aspect of the embodiment of the present application, in a third possible implementation manner of the first aspect of the embodiment of the present application, performing model training on the initial model through a plurality of mesh images further includes:
calculating a second potential energy through a Markov random field based on the commodity category identification result except the commodity category identification result corrected by the false commodity category identification result in each grid image, and using the second potential energy as a second loss function, wherein the second potential energy is used for indicating the degree of phase difference between different commodity category identification results;
performing model training on the initial model through the plurality of grid images and the first loss function includes:
and carrying out model training on the initial model through the plurality of grid images, the first loss function and the second loss function.
With reference to the third possible implementation manner of the first aspect of the embodiment of the present application, in a fourth possible implementation manner of the first aspect of the embodiment of the present application, in the process of calculating the first potential energy and/or the second potential energy by using a markov random field, the method includes:
and calculating the first potential energy and/or the second potential energy through a first-order neighborhood system in the Markov random field, wherein the first-order neighborhood system is used for indicating the range of the directly adjacent commodity category identification result of the target commodity category identification result when calculating the potential energy.
With reference to the third possible implementation manner of the first aspect of the embodiment of the present application, in a fifth possible implementation manner of the first aspect of the embodiment of the present application, before performing model training on the initial model through a plurality of mesh images, with a target of correcting an erroneous commodity category recognition result in the input mesh image, the method further includes:
converting each commodity category identification result in the plurality of grid images into a corresponding digital identifier; alternatively, the first and second electrodes may be,
and converting each commodity category identification result in the commodity category identification result set into a corresponding digital identifier, wherein the digital identifier is used for identifying different commodity category identification results.
With reference to the first aspect of the embodiment of the present application, in a sixth possible implementation manner of the first aspect of the embodiment of the present application, after the model that has been trained is taken as a modified model, the method further includes:
acquiring a target commodity category identification result, wherein the target commodity category identification result is a commodity category identification result of different commodities contained in an identification target image;
according to the positions of different commodities contained in the target image, carrying out grid distribution on the commodity category identification results of the different commodities contained in the target image to obtain a target grid image;
inputting the target grid image into a correction model, and correcting an error commodity category identification result in the target grid image, wherein the correction model is obtained by training an initial model through an error commodity category identification result set, and the error commodity category identification result set comprises different error commodity category identification results in different grid images;
and extracting a correction result of the target grid image output by the correction model.
In a second aspect, an embodiment of the present application provides an apparatus for building a correction model, where the apparatus includes:
the system comprises an acquisition unit, a processing unit and a display unit, wherein the acquisition unit is used for acquiring a commodity category identification result set, the commodity category identification result set comprises commodity category identification results of commodities contained in different images to be identified, and the commodity category identification results comprise marked wrong commodity category identification results and commodity category identification results after the wrong commodity category identification results are corrected;
the distribution unit is used for sequentially carrying out grid distribution on the commodity type identification result of the commodity contained in each image to be identified according to the placement position of the commodity to obtain a plurality of grid images;
and the training unit is used for carrying out model training on the initial model through a plurality of grid images by taking the identification result of the wrong commodity category in the input grid image as a target and taking the trained model as a correction model.
With reference to the second aspect of the embodiment of the present application, in a first possible implementation manner of the second aspect of the embodiment of the present application, the distribution unit is specifically configured to:
sequentially extracting commodity type labels from the commodity type identification result of commodities contained in each image to be identified;
and carrying out grid distribution on the multiple groups of commodity category labels according to the placing positions of the commodities to obtain multiple grid images.
With reference to the second aspect of the embodiment of the present application, in a second possible implementation manner of the second aspect of the embodiment of the present application, the training unit is specifically configured to:
calculating a first potential energy through a Markov random field based on the corrected commodity category identification result of the false commodity category identification result in each grid image, wherein the first potential energy is used for indicating the degree of difference between different commodity category identification results;
and carrying out model training on the initial model through the plurality of grid images and the first loss function.
With reference to the second possible implementation manner of the second aspect of the embodiment of the present application, in a third possible implementation manner of the second aspect of the embodiment of the present application, the training unit is further configured to:
calculating a second potential energy through a Markov random field based on the commodity category identification result except the commodity category identification result corrected by the false commodity category identification result in each grid image, and using the second potential energy as a second loss function, wherein the second potential energy is used for indicating the degree of phase difference between different commodity category identification results;
performing model training on the initial model through the plurality of grid images and the first loss function includes:
and carrying out model training on the initial model through the plurality of grid images, the first loss function and the second loss function.
With reference to the third possible implementation manner of the second aspect of the embodiment of the present application, in a fourth possible implementation manner of the second aspect of the embodiment of the present application, the training unit is specifically configured to:
and calculating the first potential energy and/or the second potential energy through a first-order neighborhood system in the Markov random field, wherein the first-order neighborhood system is used for indicating the range of the directly adjacent commodity category identification result of the target commodity category identification result when calculating the potential energy.
With reference to the third possible implementation manner of the second aspect of the embodiment of the present application, in a fifth possible implementation manner of the second aspect of the embodiment of the present application, the apparatus further includes a conversion unit, configured to:
converting each commodity category identification result in the plurality of grid images into a corresponding digital identifier; alternatively, the first and second electrodes may be,
and converting each commodity category identification result in the commodity category identification result set into a corresponding digital identifier, wherein the digital identifier is used for identifying different commodity category identification results.
With reference to the second aspect of the embodiment of the present application, in a sixth possible implementation manner of the second aspect of the embodiment of the present application, the apparatus further includes an application unit, configured to:
acquiring a target commodity category identification result, wherein the target commodity category identification result is a commodity category identification result of different commodities contained in an identification target image;
according to the positions of different commodities contained in the target image, carrying out grid distribution on the commodity category identification results of the different commodities contained in the target image to obtain a target grid image;
inputting the target grid image into a correction model, and correcting an error commodity category identification result in the target grid image, wherein the correction model is obtained by training an initial model through an error commodity category identification result set, and the error commodity category identification result set comprises different error commodity category identification results in different grid images;
and extracting a correction result of the target grid image output by the correction model.
In a third aspect, an embodiment of the present application further provides a training device for model modification, which includes a processor and a memory, where the memory stores a computer program, and the processor executes the steps in any one of the methods provided in the embodiments of the present application when calling the computer program in the memory.
In a fourth aspect, this application further provides a computer-readable storage medium, where a plurality of instructions are stored, and the instructions are adapted to be loaded by a processor to perform the steps in any one of the methods provided by this application.
As can be seen from the above, the embodiments of the present application have the following beneficial effects:
on one hand, by training the correction model of the commodity type recognition result, when the commodity type recognition model is applied, the error recognition result in the commodity type recognition result given by the commodity type recognition model can be corrected by combining the correction model, so that the recognition accuracy of the commodity type recognition result is improved.
On the other hand, the commodity type recognition results are subjected to grid distribution according to the commodity placing positions to obtain grid images, and the grid images are used for further reducing the data processing amount and improving the pertinence of the model to the commodity placing positions in the training process of the correction model, so that the training speed of the model and the correction precision of the subsequent model can be 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 flow chart illustrating a method for building a correction model according to an embodiment of the present disclosure;
FIG. 2 is a schematic diagram of a grid image according to an embodiment of the present application;
FIG. 3 is a partial schematic view of an embodiment of a product category identification result;
fig. 4 is a diagram illustrating a neighborhood system according to an embodiment of the present invention;
FIG. 5 is a flowchart illustrating step S103 according to the embodiment of FIG. 1;
FIG. 6 is a schematic diagram of a grid image after a false product category identification result is corrected according to an embodiment of the present disclosure;
FIG. 7 is a schematic structural diagram of an apparatus for building a correction model according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of a device for building a correction model according to 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 that follows, specific embodiments of the present application will be described with reference to steps and symbols executed by one or more computers, unless otherwise indicated. Accordingly, these steps and operations will be referred to, several times, as being performed by a computer, the computer performing operations involving a processing unit of the computer in electronic signals representing data in a structured form. This operation transforms the data or maintains it at locations in the computer's memory system, which may be reconfigured or otherwise altered in a manner well known to those skilled in the art. The data maintains a data structure that is a physical location of the memory that has particular characteristics defined by the data format. However, while the principles of the application have been described in language specific to above, it is not intended to be limited to the specific form set forth herein, and it will be recognized by those of ordinary skill in the art that various of the steps and operations described below may be implemented in hardware.
The principles of the present application may be employed in numerous other general-purpose or special-purpose computing, communication environments or configurations. Examples of well known computing systems, environments, and configurations that may be suitable for use with the application include, but are not limited to, hand-held telephones, personal computers, servers, multiprocessor systems, microcomputer-based systems, mainframe-based computers, and distributed computing environments that include any of the above systems or devices.
The terms "first", "second", and "third", etc. in this application are used to distinguish between different objects and not to describe a particular order. Furthermore, the terms "include" and "have," as well as any variations thereof, are intended to cover non-exclusive inclusions.
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 prior art, when commodities in a supermarket, particularly an unmanned supermarket, are monitored by an AI technology, although a commodity type identification model can be trained by combining a large number of commodity images added with commodity type labels for identifying commodities in images acquired on site, the problem of misidentification of commodities still exists, and particularly, the problem of misidentification is prominent under the conditions that actual scenes are disordered, the similarity of adjacent commodities is high, and the commodities are hidden or reflected.
Based on the above defects in the related art, the embodiments of the present application provide a method for establishing a correction model, which overcomes the defects in the related art at least to some extent.
In the method for establishing a correction model according to the embodiment of the present application, an execution main body of the method may be an establishment apparatus of the correction model, or different types of establishment apparatuses of the correction model, such as a server device, a physical host, or a User Equipment (UE), which are integrated with the establishment apparatus of the correction model, where the establishment apparatus of the correction model may be implemented in a hardware or software manner, 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), and the establishment apparatus of the correction model may be further divided into multiple apparatuses and jointly execute the establishment method of the correction model according to the embodiment of the present application.
Fig. 1 shows a schematic flow chart of a method for establishing a correction model in an embodiment of the present application, and as shown in fig. 1, the method for establishing a correction model in an embodiment of the present application may specifically include the following steps:
step S101, a commodity category identification result set is obtained, wherein the commodity category identification result set comprises commodity category identification results of commodities contained in different images to be identified, and the commodity category identification results comprise marked wrong commodity category identification results and commodity category identification results after the wrong commodity category identification results are corrected;
step S102, sequentially carrying out grid distribution on the commodity type identification result of the commodity contained in each image to be identified according to the placement position of the commodity to obtain a plurality of grid images;
step S103, with the identification result of the wrong commodity category in the input corrected grid image as the target, model training is carried out on the initial model through a plurality of grid images, and the trained model is used as a corrected model.
In the technical solution proposed in the embodiment shown in fig. 1, on one hand, by training the correction model of the product type recognition result, when the product type recognition model is applied, the error recognition result in the product type recognition result given by the product type recognition model can be corrected by combining the correction model, so as to improve the recognition accuracy of the product type recognition result.
On the other hand, the commodity type recognition results are subjected to grid distribution according to the commodity placing positions to obtain grid images, and the grid images are used for further reducing the data processing amount and improving the pertinence of the model to the commodity placing positions in the training process of the correction model, so that the training speed of the model and the correction precision of the subsequent model can be improved.
The following proceeds to a detailed description of the various steps of the embodiment shown in fig. 1:
in the embodiment of the application, the correct commodity category identification result can be the original commodity category identification result which is identified by the existing commodity category identification model and is confirmed to be correct by the staff, and a correct mark can be added for indicating that the specific commodity category identification result is correct;
the wrong product type identification result may be a product type identification result identified by a product type identification model for which the worker confirms that there is a mistake, and a mistake mark may be added to indicate that the specific product type identification result is an identification mistake. In addition, for the wrong product type recognition result, in the product type recognition result set, the corresponding product type recognition result corrected by the staff can be configured, so that the training of the subsequent correction model is facilitated.
The different product category identification results in the product category identification result set may be directly assigned with a corresponding product category label for each product on the basis of the original image, where the product category label is used to identify the product category corresponding to the product.
If the product type label in the product type identification result is embedded in the original image in the form of an image, as shown in fig. 3, which is a partial schematic view of the product type identification result according to the embodiment of the present application, the display position of the product type label may be set at a shelf on which a product is placed, and the shelf may be covered or partially covered.
At this time, because the original image acquired at the current location is affected by actual conditions, such as limited shooting space, limited shooting angle, and the like, a certain distortion and inclination still often exist, for this reason, a grid distribution process is introduced in the embodiment of the present application, that is, a commodity category identification result of a commodity included in an image to be identified is subjected to grid distribution according to a placement position of the commodity to obtain a corresponding grid image, for example, refer to a schematic diagram of the grid image of the embodiment of the present application shown in fig. 2, as shown in fig. 2, after the grid distribution process is performed, the distribution arrangement of the commodity category identification result is more orderly and normalized, and has a more obvious commodity placement position feature, so that the training of a subsequent model can be more convenient, the data processing amount is reduced, and the pertinence of the model to the placement position of the commodity is improved.
The grid distribution refers to mapping each grid according to the placing position of each commodity in the original image and the size of the space occupied by the commodity to form a grid image, and endowing a commodity type identification result for the commodity corresponding to each grid in the image. For the commodities displayed at the similar positions, the commodities are often the same or similar, and the occupied spaces are also the same or similar, so that the transformed grids can be the same, as shown in fig. 2, the grids are arranged in the form of squares with the same size; for the commodities with messy commodity placement or different commodity types, the size of the grid obtained by conversion is determined according to the placement condition of the commodities in the original image.
As an exemplary embodiment, the grid distribution process may specifically target the item category labels, that is, the item category labels may be sequentially extracted from the item category identification result of the items included in each image to be identified; and carrying out grid distribution on the multiple groups of commodity category labels according to the placing positions of the commodities to obtain multiple grid images.
If the product type label in the product type identification result is embedded in the original image in an image mode, the product type label in the current image can be extracted and subjected to grid distribution processing.
Because the grid image only contains the commodity type label and does not contain the commodity image, the method is easy to understand, and under the arrangement, the training efficiency and the training effect of a subsequent model can be obviously improved, namely the model has a correction effect with higher precision.
Further, to facilitate the training of the subsequent model, as another exemplary embodiment, the item category label (item category identification result) in fig. 2 exists in a digital form, which is easy to understand, and the item category label in the digital form also facilitates the processing on the data in the data processing.
Correspondingly, for the grid images, each commodity category identification result in the plurality of grid images can be converted into a corresponding digital identifier; alternatively, the first and second electrodes may be,
for the commodity category identification result set before the grid image is obtained, each commodity category identification result in the commodity category identification result set can be converted into a corresponding digital identifier, and the digital identifiers are used for identifying different commodity category identification results.
In the embodiment of the application, the initial model is an initialized neural network model, the model comprises an input layer, a convolutional layer and a full-connection layer, the input layer converts an input image into an image vector, the convolutional layer extracts different feature vectors from the input image vector, and the full-connection layer classifies the feature vectors output by the convolutional layer and maps the feature vectors to a sample mark space to serve as a recognition result of the model.
In the training process of the model, different grid images can be sequentially input into the model to carry out forward propagation, then a loss function is calculated according to a correction result output by the model, and backward propagation is carried out, so that the parameter adjustment and optimization are carried out on the model, and the model after training is a correction model which can be put into practical application after repeated propagation and optimization, so that the establishment of the correction model is completed.
In the training process, as another exemplary embodiment, a markov random field may be specifically combined as a basis for identifying a false commodity category identification result, wherein, in turn, a calculation of potential energy may be involved, the potential energy being used to indicate a degree of phase difference between different commodity category identification results, the larger the potential energy, the larger the phase difference, the more likely it is not to belong to the same commodity, and the smaller the potential energy, the more likely it is to belong to the same commodity.
The specific calculation process of the potential energy can specifically refer to the following contents:
after the commodity category identification result is converted into the grid image, a Maximum A Posteriori (MAP) algorithm can be adopted to convert the correction problem of the identification result into a MAP problem for solving the image.
The MAP estimator may be described as:
In the formula 2, the first and second groups,
wherein X is the required item class label, called PX(ω) is the prior probability of the marker ω.
For the denominator portion of equation 2, PF(f) Is the probability of the observed value f, since the observed data f is given, so PF(f) Is a constant. Therefore, the denominator part in equation 2 is a value that can be determined in the subsequent calculation.
For the first half of the molecule of equation 2, PF|X(F, ω) is a conditional probability density function (also called likelihood function) of the observed value F, which is a probability description of obtaining the observed image F from the marker image X. According to the theorem of large numbers, a probability density function P (f) is assumeds|ωs) Obeying a Gaussian distribution, its mean value μ can be usedλAnd σλTo show the distribution rule. The likelihood energy function can thus be expressed as follows:
in the formula 3, the first and second phases,
for the second half of the numerator of equation 2, i.e., the prior probability PX(ω), assuming it is a markov random field, the key to solving is to define its potential energy of the potential mass. In consideration of computational efficiency, a first-order neighborhood (3-neighborhood) and a second-order neighborhood (8-neighborhood) are generally used in the markov model. The first and second order neighborhood systems and their corresponding radicals can refer to a schematic diagram of the neighborhood system of the embodiment of the present application shown in fig. 4.
The first-order neighborhood system is used for indicating the range of the commodity category identification result directly adjacent to the target commodity category identification result when the potential energy is calculated.
As still another exemplary embodiment, in consideration of a general rule of shelf goods placement, the correlation between two horizontally adjacent goods is the largest, and the correlation between vertically adjacent goods is generally weak, i.e., the correlation between a goods adjacent in a 45 ° direction, a goods adjacent in a 135 ° direction, and three or more adjacent goods is weak. Therefore, only a first-order neighborhood system can be selected, and the influence of the commodity class consistency in the horizontal direction on the energy function is enhanced.
The first order radical potential can be expressed as follows:
in the case of the formula 4,
wherein, ω issAnd ωrAre a class of two contiguous commodities in a first-order radical. Alpha and beta are model parameters, and the value is 0 according to the actual situation<α<β<1.0, they control the isomorphism of the area, namely the punishment quantity when the prediction categories of two commodities adjacent to each other are different in the commodity identification result correction. The single-point radicals in the first-order neighborhood system are not considered in the above equation because for a single isolated commodity, its surrounding commodities are either vacancies (labeled-1) or other commodities whose double-point radical potential must be positive. That is, single-point radicals may be reflected in a double-point radical potential.
After defining the potential of the radical, the second half of the molecule of equation 2, i.e., the prior probability P, can be definedXThe energy function of (ω) is:
The posterior energy can be expressed as:
U(ω,f)＝U1(ω,f)+U2(ω)
in this way, the calculated U (ω, f) is used as the potential energy of the commodity category identification result.
In the specific process of model training according to potential energy, the corresponding potential energy can be calculated according to the corrected commodity category identification result of the wrong commodity category identification result in each grid image, so that when the commodity category identification result is corrected, the size range of the potential energy of the corrected commodity category identification result is determined, the potential energy is used as a loss function for model training, specifically, iterative solution can be carried out by using methods such as simulated annealing, decisive annealing and the like, and therefore, a more accurate commodity category identification result which is more in line with the goods shelf commodity arrangement rule is obtained.
Correspondingly, step S103 in the embodiment corresponding to fig. 1 is described above, and as shown in fig. 5, a schematic flow chart of step S103 in the embodiment corresponding to fig. 1 of the present application may include:
step S501, calculating a first potential energy through a Markov random field based on the corrected commodity category identification result of the false commodity category identification result in each grid image, and using the first potential energy as a first loss function, wherein the first potential energy is used for indicating the degree of phase difference between different commodity category identification results;
step S502, model training is carried out on the initial model through the grid images and the first loss function.
It can be understood that the potential energy obtained by combining the Markov random field calculation is used as a loss function, and a more powerful data support can be further provided for identifying and correcting the false commodity category identification result, and in the correction process, the connectivity of adjacent commodities is enhanced, so that the correction accuracy of the model is continuously and effectively improved on the basis of the model.
Referring to fig. 6, a schematic diagram of a grid image after correcting a false product category identification result according to the embodiment of the present invention, in the original image 2, the initial product category identification result is different from the product category identification categories of the right and left adjacent products and also different from the product category identification categories of the upper and lower adjacent products, in the 3 rd row, the 4 th row and the 5 th row, the 2 nd row, and therefore, the initial product category identification result is corrected in the process of optimizing the posterior energy, and becomes the correct product category identification result shown in fig. 5.
Furthermore, in addition to considering the wrong commodity category identification result, the potential energy of all the commodity category identification results can be considered in the overall aspect to train the model. Correspondingly, step S103 in the corresponding embodiment in fig. 1 may further include:
calculating a second potential energy by a Markov random field based on the commodity category identification result other than the commodity category identification result corrected by the false commodity category identification result in each grid image;
meanwhile, step S502 in the embodiment corresponding to fig. 5 may include:
and carrying out model training on the initial model through the plurality of grid images, the first loss function and the second loss function.
Under the arrangement, based on the overall layer, the contact of adjacent commodities is strengthened again, the correction result of the correction model is strived to be obtained, the potential energy is the lowest, and the identification precision of the commodity category is guaranteed at the maximum probability.
After the training of the model is completed, the obtained correction model can be put into practical application to realize the correction method of the commodity category identification result, and correspondingly, the application can be realized through the following steps:
acquiring a target commodity category identification result, wherein the target commodity category identification result is a commodity category identification result of different commodities contained in an identification target image;
according to the positions of different commodities contained in the target image, carrying out grid distribution on the commodity category identification results of the different commodities contained in the target image to obtain a target grid image;
inputting the target grid image into a correction model, and correcting an error commodity category identification result in the target grid image, wherein the correction model is obtained by training an initial model through an error commodity category identification result set, and the error commodity category identification result set comprises different error commodity category identification results in different grid images;
and extracting a correction result of the target grid image output by the correction model.
The modification model established by the method for establishing a modification model according to the embodiment of the present application may be independent or embedded in an existing product type recognition model, so that the recognition accuracy of the product type recognition model is further improved by combining the modification process of the modification model.
Similarly, the method for establishing the correction model provided in the embodiment of the present application may also be embedded in the process of establishing the existing product category identification model, so that the obtained product category identification model is established, and may also have a function due to the correction model established by the method for establishing the correction model provided in the embodiment of the present application.
In order to better implement the training method of the correction model provided in the embodiment of the present application, the embodiment of the present application further provides a device for establishing the correction model.
Referring to fig. 7, fig. 7 is a schematic structural diagram of an apparatus for building a correction model according to an embodiment of the present disclosure, in which the apparatus 700 for building a correction model specifically includes the following structures:
an obtaining unit 701, configured to obtain a product category identification result set, where the product category identification result set includes product category identification results of products included in different images to be identified, and the product category identification result includes a marked incorrect product category identification result and a product category identification result after the incorrect product category identification result is corrected;
a distribution unit 702, configured to sequentially perform grid distribution on the product category identification result of the product included in each image to be identified according to the placement position of the product, so as to obtain a plurality of grid images;
a training unit 703 is configured to perform model training on the initial model through a plurality of mesh images with the target of correcting the erroneous commodity category recognition result in the input mesh image, and use the trained model as a correction model.
As an exemplary embodiment, the distribution unit 702 is specifically configured to:
sequentially extracting commodity type labels from the commodity type identification result of commodities contained in each image to be identified;
and carrying out grid distribution on the multiple groups of commodity category labels according to the placing positions of the commodities to obtain multiple grid images.
As another exemplary embodiment, the training unit 703 is specifically configured to:
calculating a first potential energy through a Markov random field based on the corrected commodity category identification result of the false commodity category identification result in each grid image, wherein the first potential energy is used for indicating the degree of difference between different commodity category identification results;
and carrying out model training on the initial model through the plurality of grid images and the first loss function.
As another exemplary embodiment, the training unit 703 is further configured to:
calculating a second potential energy through a Markov random field based on the commodity category identification result except the commodity category identification result corrected by the false commodity category identification result in each grid image, and using the second potential energy as a second loss function, wherein the second potential energy is used for indicating the degree of phase difference between different commodity category identification results;
performing model training on the initial model through the plurality of grid images and the first loss function includes:
and carrying out model training on the initial model through the plurality of grid images, the first loss function and the second loss function.
As another exemplary embodiment, the training unit 703 is specifically configured to:
and calculating the first potential energy and/or the second potential energy through a first-order neighborhood system in the Markov random field, wherein the first-order neighborhood system is used for indicating the range of the directly adjacent commodity category identification result of the target commodity category identification result when calculating the potential energy.
As a further exemplary embodiment, the apparatus further comprises a conversion unit 704 for:
converting each commodity category identification result in the plurality of grid images into a corresponding digital identifier; alternatively, the first and second electrodes may be,
and converting each commodity category identification result in the commodity category identification result set into a corresponding digital identifier, wherein the digital identifier is used for identifying different commodity category identification results.
As a further exemplary embodiment, the apparatus further comprises an applying unit 705 for:
acquiring a target commodity category identification result, wherein the target commodity category identification result is a commodity category identification result of different commodities contained in an identification target image;
according to the positions of different commodities contained in the target image, carrying out grid distribution on the commodity category identification results of the different commodities contained in the target image to obtain a target grid image;
inputting the target grid image into a correction model, and correcting an error commodity category identification result in the target grid image, wherein the correction model is obtained by training an initial model through an error commodity category identification result set, and the error commodity category identification result set comprises different error commodity category identification results in different grid images;
and extracting a correction result of the target grid image output by the correction model.
Referring to fig. 8, fig. 8 shows a schematic structural diagram of the device for establishing a correction model according to the embodiment of the present application, specifically, the device for establishing a correction model according to the present application includes a processor 801, where the processor 801 is configured to implement, when executing a computer program stored in a memory 802, each step of the method for establishing a correction model according to any embodiment corresponding to fig. 1 to 6; alternatively, the processor 801 is configured to implement the functions of the units in the corresponding embodiment of fig. 7 when executing the computer program stored in the memory 802.
Illustratively, a computer program may be partitioned into one or more modules/units, which are stored in the memory 802 and executed by the processor 801 to accomplish 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 device for creating the correction model may include, but is not limited to, the processor 801 and the memory 802. It will be understood by those skilled in the art that the illustration is merely an example of the device for building the correction model, and does not constitute a limitation of the device for building the correction model, and may include more or less components than those shown, or combine some components, or different components, for example, the device for building the correction model may further include an input-output device, a network access device, a bus, etc., and the processor 801, the memory 802, the input-output device, the network access device, etc. are connected through the bus.
The Processor 801 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, discrete hardware components, etc. The general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor being the control center of the model-modifying building apparatus, with various interfaces and lines connecting the various parts of the overall apparatus.
The memory 802 may be used to store computer programs and/or modules, and the processor 801 may implement various functions of the computer device by running or executing the computer programs and/or modules stored in the memory 802 and invoking data stored in the memory 802. The memory 802 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, etc.) created according to the use of the creation device of the revision 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 device, the apparatus, and the corresponding units for establishing the correction model described above may refer to the descriptions of the method in any embodiment corresponding to fig. 1 to fig. 6, 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.
To this end, 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 the method for establishing a correction model in any embodiment of the present application corresponding to fig. 1 to 6, and specific operations may refer to descriptions of the method for establishing a correction model in any embodiment corresponding to fig. 1 to 6, which are not described herein again.
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 establishing the correction model in any embodiment of the present application, such as those shown in fig. 1 to fig. 6, the beneficial effects that can be achieved by the method for establishing the correction model in any embodiment of the present application, such as those shown in fig. 1 to fig. 6, can be achieved, which are detailed in the foregoing description and will not be repeated herein.
The method, the apparatus, the device and the computer-readable storage medium for establishing the correction model provided by the present application are described in detail above, 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 idea 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 method for building a correction model, the method comprising:
acquiring a commodity category identification result set, wherein the commodity category identification result set comprises commodity category identification results of commodities contained in different images to be identified, and the commodity category identification results comprise marked wrong commodity category identification results and commodity category identification results after the wrong commodity category identification results are corrected;
sequentially carrying out grid distribution on the commodity category identification result of the commodity contained in each image to be identified according to the placement position of the commodity to obtain a plurality of grid images;
and performing model training on the initial model through a plurality of grid images by taking the error commodity category identification result in the input grid image as a target, and taking the trained model as a correction model.
2. The method according to claim 1, wherein the sequentially performing grid distribution on the product category identification result of the product included in each image to be identified according to the placement position of the product to obtain a plurality of grid images comprises:
sequentially extracting commodity type labels from the commodity type identification result of commodities contained in each image to be identified;
and carrying out grid distribution on the multiple groups of commodity category labels according to the placing positions of the commodities to obtain multiple grid images.
3. The method of claim 1, wherein the model training of the initial model over the plurality of mesh images comprises:
calculating a first potential energy through a Markov random field based on the corrected commodity category identification result of the false commodity category identification result in each grid image, wherein the first potential energy is used for indicating the degree of phase difference between different commodity category identification results and is used as a first loss function;
and performing model training on the initial model through a plurality of grid images and the first loss function.
4. The method of claim 3, wherein model training the initial model through the plurality of mesh images further comprises:
calculating a second potential energy through the Markov random field based on the commodity category identification result except the commodity category identification result corrected by the false commodity category identification result in each grid image, wherein the second potential energy is used for indicating the degree of phase difference between different commodity category identification results;
model training an initial model through the plurality of mesh images and the first loss function includes:
and performing model training on the initial model through a plurality of grid images, the first loss function and the second loss function.
5. The method of claim 4, wherein calculating the first potential energy and/or the second potential energy using a Markov random field comprises:
and calculating the first potential energy and/or the second potential energy through a first-order neighborhood system in the Markov random field, wherein the first-order neighborhood system is used for indicating a range of the commodity category identification results which are directly adjacent to the target commodity category identification result when the potential energy is calculated.
6. The method according to claim 4, wherein before the model training of the initial model by the plurality of mesh images with the aim of correcting the erroneous commodity category recognition result in the input mesh image, the method further comprises:
converting each commodity category identification result in the plurality of grid images into a corresponding digital identifier; alternatively, the first and second electrodes may be,
and converting each commodity category identification result in the commodity category identification result set into a corresponding digital identifier, wherein the digital identifier is used for identifying different commodity category identification results.
7. The method of claim 1, wherein after the training of the model is completed as a modified model, the method further comprises:
acquiring a target commodity category identification result, wherein the target commodity category identification result is a commodity category identification result for identifying different commodities contained in a target image;
according to the positions of different commodities contained in the target image, carrying out grid distribution on the commodity category identification results of the different commodities contained in the target image to obtain a target grid image;
inputting the target grid image into the correction model, and correcting an error commodity category identification result in the target grid image, wherein the correction model is obtained by training an initial model through an error commodity category identification result set, and the error commodity category identification result set comprises different error commodity category identification results in different grid images;
and extracting a correction result of the target grid image output by the correction model.
8. An apparatus for creating a correction model, the apparatus comprising:
the system comprises an acquisition unit, a processing unit and a display unit, wherein the acquisition unit is used for acquiring a commodity category identification result set, the commodity category identification result set comprises commodity category identification results of commodities contained in different images to be identified, and the commodity category identification results comprise marked wrong commodity category identification results and commodity category identification results after the wrong commodity category identification results are corrected;
the distribution unit is used for sequentially carrying out grid distribution on the commodity type identification result of the commodity contained in each image to be identified according to the placement position of the commodity to obtain a plurality of grid images;
and the training unit is used for carrying out model training on the initial model through a plurality of grid images by taking the identification result of the wrong commodity category in the input grid image as a target and taking the trained model as a correction model.
9. An apparatus for building a correction model, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the method according to any one of claims 1 to 7 when calling the computer program in the memory.
10. A computer-readable storage medium storing a plurality of instructions adapted to be loaded by a processor to perform the method of any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010147659.5A CN113361511A (en) | 2020-03-05 | 2020-03-05 | Method, device and equipment for establishing correction model and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010147659.5A CN113361511A (en) | 2020-03-05 | 2020-03-05 | Method, device and equipment for establishing correction model and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113361511A true CN113361511A (en) | 2021-09-07 |
Family
ID=77523831
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010147659.5A Pending CN113361511A (en) | 2020-03-05 | 2020-03-05 | Method, device and equipment for establishing correction model and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113361511A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115063564A (en) * | 2022-07-13 | 2022-09-16 | 如你所视(北京)科技有限公司 | Article label display method, device and medium for two-dimensional display image |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012111104A1 (en) * | 2011-02-16 | 2012-08-23 | 株式会社日立製作所 | Analysis model creation assistance device and analysis model creation assistance program |
| CN103489009A (en) * | 2013-09-17 | 2014-01-01 | 北方信息控制集团有限公司 | Pattern recognition method based on self-adaptation correction neural network |
| CN107239203A (en) * | 2016-03-29 | 2017-10-10 | 北京三星通信技术研究有限公司 | A kind of image management method and device |
| CN108009589A (en) * | 2017-12-12 | 2018-05-08 | 腾讯科技（深圳）有限公司 | Sample data processing method, device and computer-readable recording medium |
| CN108595410A (en) * | 2018-03-19 | 2018-09-28 | 小船出海教育科技（北京）有限公司 | The automatic of hand-written composition corrects method and device |
| CN108875821A (en) * | 2018-06-08 | 2018-11-23 | Oppo广东移动通信有限公司 | The training method and device of disaggregated model, mobile terminal, readable storage medium storing program for executing |
| CN109313710A (en) * | 2018-02-02 | 2019-02-05 | 深圳蓝胖子机器人有限公司 | Model of Target Recognition training method, target identification method, equipment and robot |
| CN109472712A (en) * | 2018-10-12 | 2019-03-15 | 天津大学 | A kind of efficient Markov random field Combo discovering method strengthened based on structure feature |
| CN109558600A (en) * | 2018-11-14 | 2019-04-02 | 北京字节跳动网络技术有限公司 | Translation processing method and device |
| CN109977251A (en) * | 2019-03-05 | 2019-07-05 | 武汉摩小超科技有限公司 | A method of building identifies commodity based on RGB histogram feature |
| CN110414532A (en) * | 2019-06-21 | 2019-11-05 | 广州利科科技有限公司 | A kind of commodity image recognition methods based on deep learning |
| CN110503143A (en) * | 2019-08-14 | 2019-11-26 | 平安科技（深圳）有限公司 | Research on threshold selection, equipment, storage medium and device based on intention assessment |
| CN110610149A (en) * | 2019-09-03 | 2019-12-24 | 卓尔智联(武汉)研究院有限公司 | Information processing method and device and computer storage medium |
| CN110766062A (en) * | 2019-10-15 | 2020-02-07 | 广州织点智能科技有限公司 | Commodity recognition model training method and device, electronic equipment and storage medium |
-
2020
- 2020-03-05 CN CN202010147659.5A patent/CN113361511A/en active Pending
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012111104A1 (en) * | 2011-02-16 | 2012-08-23 | 株式会社日立製作所 | Analysis model creation assistance device and analysis model creation assistance program |
| CN103489009A (en) * | 2013-09-17 | 2014-01-01 | 北方信息控制集团有限公司 | Pattern recognition method based on self-adaptation correction neural network |
| CN107239203A (en) * | 2016-03-29 | 2017-10-10 | 北京三星通信技术研究有限公司 | A kind of image management method and device |
| CN108009589A (en) * | 2017-12-12 | 2018-05-08 | 腾讯科技（深圳）有限公司 | Sample data processing method, device and computer-readable recording medium |
| CN109313710A (en) * | 2018-02-02 | 2019-02-05 | 深圳蓝胖子机器人有限公司 | Model of Target Recognition training method, target identification method, equipment and robot |
| CN108595410A (en) * | 2018-03-19 | 2018-09-28 | 小船出海教育科技（北京）有限公司 | The automatic of hand-written composition corrects method and device |
| CN108875821A (en) * | 2018-06-08 | 2018-11-23 | Oppo广东移动通信有限公司 | The training method and device of disaggregated model, mobile terminal, readable storage medium storing program for executing |
| CN109472712A (en) * | 2018-10-12 | 2019-03-15 | 天津大学 | A kind of efficient Markov random field Combo discovering method strengthened based on structure feature |
| CN109558600A (en) * | 2018-11-14 | 2019-04-02 | 北京字节跳动网络技术有限公司 | Translation processing method and device |
| CN109977251A (en) * | 2019-03-05 | 2019-07-05 | 武汉摩小超科技有限公司 | A method of building identifies commodity based on RGB histogram feature |
| CN110414532A (en) * | 2019-06-21 | 2019-11-05 | 广州利科科技有限公司 | A kind of commodity image recognition methods based on deep learning |
| CN110503143A (en) * | 2019-08-14 | 2019-11-26 | 平安科技（深圳）有限公司 | Research on threshold selection, equipment, storage medium and device based on intention assessment |
| CN110610149A (en) * | 2019-09-03 | 2019-12-24 | 卓尔智联(武汉)研究院有限公司 | Information processing method and device and computer storage medium |
| CN110766062A (en) * | 2019-10-15 | 2020-02-07 | 广州织点智能科技有限公司 | Commodity recognition model training method and device, electronic equipment and storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115063564A (en) * | 2022-07-13 | 2022-09-16 | 如你所视(北京)科技有限公司 | Article label display method, device and medium for two-dimensional display image |
| CN115063564B (en) * | 2022-07-13 | 2024-04-30 | 如你所视(北京)科技有限公司 | Method, device and medium for displaying object label in two-dimensional display image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10936911B2 (en) | Logo detection | |
| US11670071B2 (en) | Fine-grained image recognition | |
| US10762373B2 (en) | Image recognition method and device | |
| US8442307B1 (en) | Appearance augmented 3-D point clouds for trajectory and camera localization | |
| US20140010407A1 (en) | Image-based localization | |
| CN105164700B (en) | Detecting objects in visual data using a probabilistic model | |
| CN109816769A (en) | Scene based on depth camera ground drawing generating method, device and equipment | |
| CN103927387A (en) | Image retrieval system, method and device | |
| CN111860398B (en) | Remote sensing image target detection method and system and terminal equipment | |
| CN111680678B (en) | Target area identification method, device, equipment and readable storage medium | |
| CN110852349A (en) | Image processing method, detection method, related equipment and storage medium | |
| CN111753863A (en) | Image classification method and device, electronic equipment and storage medium | |
| CN108229680B (en) | Neural network system, remote sensing image recognition method, device, equipment and medium | |
| CN110942473A (en) | Moving target tracking detection method based on characteristic point gridding matching | |
| CN111008589B (en) | Face key point detection method, medium, device and computing equipment | |
| CN111340213B (en) | Neural network training method, electronic device, and storage medium | |
| CN111368860B (en) | Repositioning method and terminal equipment | |
| CN109961516B (en) | Surface acquisition method, device and non-transitory computer readable recording medium | |
| CN114168768A (en) | Image retrieval method and related equipment | |
| CN113361511A (en) | Method, device and equipment for establishing correction model and computer readable storage medium | |
| CN112861934A (en) | Image classification method and device of embedded terminal and embedded terminal | |
| CN112364821A (en) | Self-recognition method and device for power mode data of relay protection device | |
| CN112036362A (en) | Image processing method, image processing device, computer equipment and readable storage medium | |
| CN111104965A (en) | Vehicle target identification method and device | |
| CN116309643A (en) | Face shielding score determining method, 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 |