KR102076075B1 - Method to generate data - Google Patents

Abstract

Disclosed is a computer program stored in a computer readable storage medium. The computer program, when executed on one or more processors of a computing device, performs an operation for providing a method for generating learning data, and the operation comprises: identifying an object included in an image based on the image included in the first learning data; modifying the image based on at least a part of the object; and generating the second learning data including the modified image.

Description

How to generate data {METHOD TO GENERATE DATA}

The present invention relates to a method for generating data, and more particularly, to a method for generating training data.

In order to classify data through machine learning, securing learning data for machine learning must be preceded.

In constructing the training data, there is a problem in that an imbalance occurs in each class of data. For example, the number of data corresponding to the defective semiconductor in the semiconductor process may be less than the number of data corresponding to the normal semiconductor corresponding to the semiconductor having no defect. If the learning is performed through a set of skewed data in one class with such a class imbalance, the resulting network function model may be less accurate.

Accordingly, there is a need in the art to solve the imbalance problem of data in constructing the learning data.

Republic of Korea Patent Publication No. 2016-0012537 discloses a neural network learning method and apparatus, data processing apparatus.

The present disclosure has been made in response to the above-described background, and an object thereof is to provide a data generation method.

A computer program stored in a computer readable storage medium according to an embodiment of the present disclosure for realizing the problems as described above, wherein the computer program is based on an image contained in the first learning data, the object contained in the image Identifying; Modifying an image based on at least a portion of the object; And generating second training data including the modified image.

In an alternate embodiment of computer program operations for performing the following operations for generating training data, identifying an object included in the image based on the image included in the first training data is included in the image. May include identifying an object of interest.

In an alternative embodiment of computer program operations to perform the following operations for learning data generation, the object of interest may comprise a labeled object.

In an alternative embodiment of computer program operations to perform the following operations for generating training data, the object of interest may comprise anomaly data.

In an alternate embodiment of computer program operations for performing the following operations for generating training data, identifying an object included in the image based on the image included in the first training data is included in the image. And generating a segmentation of the segmented object.

In an alternate embodiment of computer program operations to perform the following operations for generating training data, modifying the image based on at least a portion of the object crops the image to include at least a portion of the object. It may include an operation of cropping.

In an alternative embodiment of computer program operations for performing the following operations for generating training data, cropping the image to include at least a portion of the object may include: cropping the first portion included in the object. Cropping the image to be included in the pinged image.

In an alternative embodiment of computer program operations to perform the following operations for generating training data, cropping the image to include at least a portion of the object includes: first cropping the object and the cropping portion; Cropping the image such that the second portion of the pinged image overlaps.

In an alternative embodiment of computer program operations to perform the following operations for generating training data, the second portion may be included in an area within a predetermined range at the center of the cropped image.

In an alternative embodiment of computer program operations to perform the following operations for generating training data, modifying an image based on at least a portion of the object may modify a portion of the image except at least a portion of the object. May include an action.

In an alternative embodiment of computer program operations for performing the following operations for generating training data, modifying an image based on at least a portion of the object comprises: masking a portion other than at least a portion of the object. It may include.

In an alternate embodiment of computer program operations to perform the following operations for generating training data, generating an training data set comprising the first training data and the second training data; And training a model including one or more network functions to identify an object included in an image input to the model based on the training data set.

A method of generating training data according to an embodiment of the present disclosure for realizing an object as described above, the method comprising: identifying an object included in the image based on an image included in first training data; Modifying an image based on at least a portion of the object; And generating second training data including the modified image.

A server for providing a method of generating training data according to an embodiment of the present disclosure for realizing the above-mentioned problem, the server comprising: a processor including one or more cores; And a memory, wherein the processor is further configured to: identify an object included in the image based on the image included in the first training data, modify the image based on at least a portion of the object, and modify the modified image. The second training data may be generated.

The present disclosure can provide a data generation method.

1 is a block diagram of a computing device performing an operation for providing a method of generating training data according to an embodiment of the present disclosure.
2A, 2B, and 2C illustrate examples of a method of generating training data according to an embodiment of the present disclosure.
3 is a diagram illustrating an example of a method of generating training data according to an embodiment of the present disclosure.
4 is a flowchart of a method of generating training data according to an embodiment of the present disclosure.
5 is a block diagram of a computing device according to an embodiment of the present disclosure.

Various embodiments are now described with reference to the drawings. In the present specification, various descriptions are presented to provide an understanding of the present disclosure. It may be evident, however, that such embodiments may be practiced without these specific details.

As used herein, the terms “component”, “module”, “system” and the like refer to computer-related entities, hardware, firmware, software, a combination of software and hardware, or the execution of software. For example, a component may be, but is not limited to being, a process running on a processor, a processor, an object, a thread of execution, a program, and / or a computer. For example, both an application running on a computing device and the computing device can be a component. One or more components can reside within a processor and / or thread of execution. One component can be localized within one computer. One component may be distributed between two or more computers. In addition, these components can execute from various computer readable media having various data structures stored thereon. The components may be connected via a network such as the Internet and other systems via, for example, signals with one or more data packets (e.g., data and / or signals from one component that interacts with other components in a local system, distributed system). Data may be communicated via local and / or remote processes.

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or”. That is, unless specified otherwise or unambiguously in context, "X uses A or B" is intended to mean one of the natural implicit substitutions. That is, X uses A; X uses B; Or where X uses both A and B, "X uses A or B" may apply in either of these cases. In addition, the term "and / or" as used herein is to be understood to refer to and include all possible combinations of one or more of the listed related items.

In addition, the terms "comprises" and / or "comprising" should be understood to mean that the corresponding features and / or components are present. It is to be understood, however, that the terms "comprises" and / or "comprising" do not exclude the presence or addition of one or more other features, components, and / or groups thereof. Also, unless otherwise specified or in the context of indicating a singular form, the singular in the specification and claims should generally be construed as meaning "one or more."

Those skilled in the art will further appreciate that the various illustrative logical blocks, configurations, modules, circuits, means, logics, and algorithm steps described in connection with the embodiments disclosed herein may be electronic hardware, computer software, or a combination of both. It should be appreciated that it can be implemented with To clearly illustrate the interchangeability of hardware and software, various illustrative components, blocks, configurations, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented in hardware or as software depends on the particular application and design constraints imposed on the overall system. Skilled artisans can implement the described functionality in a variety of ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be apparent to those of ordinary skill in the art. The general principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present invention is not limited to the embodiments presented herein. The invention should be construed in the broadest scope consistent with the principles and novel features set forth herein.

In one embodiment of the present disclosure, the server may include other configurations for performing the server environment of the server. The server may include all types of devices. The server may be a digital device, and may be a digital device having a computing power including a processor and having a memory such as a laptop computer, a notebook computer, a desktop computer, a web pad, a mobile phone, and the like. The server may be a web server that handles services. The type of server described above is merely an example and the present disclosure is not limited thereto.

1 is a block diagram of a computing device performing an operation for providing a method of generating training data according to an embodiment of the present disclosure.

The computing device 100 for providing a method of generating training data according to an embodiment of the present disclosure may include a network unit 110, a processor 120, and a memory 130.

The network unit 110 may transmit and receive an image, first training data, second training data, and the like, according to an exemplary embodiment of the present disclosure.

The processor 120 may include one or more cores, and may include a central processing unit (CPU), a general purpose graphics processing unit (GPGPU), and a tensor processing unit (TPU) of a computing device. and a processor for performing learning data generation, such as a unit). The processor 120 may read a computer program stored in the memory 130 to provide a method of generating training data according to an embodiment of the present disclosure. According to an embodiment of the present disclosure, the processor 120 may perform calculations to provide a method of generating learning data.

The memory 130 may store a computer program for providing a method of generating learning data according to an embodiment of the present disclosure, and the stored computer program may be read and driven by the processor 120.

The memory 130 according to embodiments of the present disclosure may store a program for the operation of the processor 120, and may temporarily or permanently store input / output data or events. The memory 130 may store data regarding a display and sound. The memory 130 may include a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (for example, SD or XD memory), RAM (Random Access Memory, RAM), Static Random Access Memory (SRAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Programmable Read-Only Memory (PROM), Magnetic Memory, Magnetic It may include a storage medium of at least one type of disk, optical disk.

The processor 120 may perform data augmentation to generate new training data based on the first training data. The first training data may be training data that is a basis for generating new training data. The processor 120 may generate second training data based on the first training data.

The processor 120 may perform data aggregation on a class lacking training data in a training data set for training a neural network model including one or more network functions. Training data for training the neural network model may include training data for each of two or more classes. The neural network model may be a model for classifying data.

The class imbalance may refer to a case in which there is a bias in the ratio of training data corresponding to each class. A neural network model trained with training data with class imbalances may have the problem of outputting results toward a higher proportion of training data. For example, if one percent of the training data in a disease learning data set has a positive label and the remaining 99 percent of the training data has a negative label, it is a class imbalance. For example, if 0.48 percent of the training data in the entity classification data set has the A entity label and the remaining 0.52 percent of the training data has the B entity label, then this is not a class imbalance. The training data generation method according to the present disclosure is to solve the class imbalance if there is.

Training data for some classes of the training data set that includes training data for each of the two or more classes may be insufficient. In the case where the neural network model is trained while the training data for some classes are insufficient, accurate training may be difficult for the classes lacking the training data. The processor 120 may perform data aggregation on a class lacking training data in order to perform accurate learning on all classes included in the training data set.

The processor 120 may perform data aggregation based on learning data belonging to a class lacking learning data. The processor 120 additionally adds training data belonging to a relatively insufficient class by performing data aggregation on a class including training data that includes a plurality of classes of training data. Can be generated.

For example, the processor 120 may perform defect inspection on a product in a production process. In order for the processor 120 to train a neural network model for performing defect checking on a product, a training data set may be required that includes training data based on a defective product and training data for a product without a defect. The number of defective training data may be less than the number of defective training data. For example, the number of training data based on products without defects (i.e., normal data) may be a sufficient number necessary for training the neural network model, and the training data based on defective products (i.e., normal data) The number of anomaly data may be smaller than the number necessary for training the neural network model. Specific description of the generated learning data and its class is only an example, and the present disclosure is not limited thereto.

The processor 120 may generate second training data based on the first training data. The processor 120 may generate second training data by performing data aggregation based on the first training data. The processor 120 may transform the first training data to generate second training data. For example, the deformation of the first training data may invert, randomly crop, mask, or scale at least a portion of the first image included in the first training data. Or change the saturation or brightness, add noise, rotate, and the like. The detailed description of the above-described modification is only an example and the present disclosure is not limited thereto.

The processor 120 may train the neural network model based on a training data set including the first training data and the second training data.

Hereinafter, a method of transforming the first training data will be described in detail. Hereinafter, a method of transforming the first training data will be described with reference to FIGS. 2A, 2B, and 2C. 2A, 2B and 2C illustrate examples of a method of generating training data according to an embodiment of the present disclosure. The images shown in FIGS. 2A, 2B, and 2C are merely examples for explaining a method of generating training data, and the present disclosure may include any training data and any images.

2A is a schematic diagram illustrating a method of cropping a first image 300 included in first training data.

In the present example, the first training data may include input data that can be input to the neural network model and a label indicating an attribute of the input data, and the input data may be configured, for example, as an image.

The processor 120 may identify the object 301 in the first image 300 included in the first training data. The first image 300 may be an image input to the neural network model for learning the neural network model. The first image 300 may be an image of the training data corresponding to the class lacking the training data. In the above example, the first image 300 may be an image of a defective product. In the above example, the first image 300 may be an image of a product including an anomaly (for example, a defective product). Specific description of the image included in the above-described first training data is only an example and the present disclosure is not limited thereto.

The processor 120 may identify the object 301 included in the first image 300. The processor 120 may generate new training data by converting the first image 300 based on the object 301.

Object 301 may include an object of interest. The object of interest may be a region including specific information to be acquired in the first image 300. For example, when detecting whether a defect is detected through the first image 300, the specific information may be anomaly data, and the object of interest may be an anomaly region. The object of interest may be an area to be calculated using the neural network model in the first image 300 or an area to be calculated and obtained using the neural network model.

The object of interest may include a labeled object.

The label according to an embodiment of the present disclosure may be, for example, a segmentation label. The segmentation label may indicate a partial region included in the first image 300 to be distinguished from other regions of the image. The segmentation label may be a label for displaying location information on the object 301 included in the first image 300 of the first training data. The labeled object may be displayed to distinguish the object 301 included in the first image 300 from other areas of the image. The labeled object may be displayed to distinguish pixels of an area included in the object from pixels of an area not included in the object. For example, if the object 301 is an anomaly region (eg, a portion where a defect is identified in the captured image, etc.), the labeled object is an anomaly region with respect to the anomaly region in the first image 300. It may be displayed to be distinguished from other areas except for. Specific description of the above-described labeled object is only an example and the present disclosure is not limited thereto.

The label according to another embodiment of the present disclosure may be a classification label. The classification label may indicate information on which class the first image 300 belongs to. The classification label may be displayed as to whether the first image 300 includes specific information that the user wants to acquire from the image. For example, when performing a defect test of a product obtained in a product process, the product may be classified as normal or anomaly. In the case of defect inspection on the image of the product obtained in the product process, there may be classes for each of normal and anomaly. An image containing anomaly data may be a label for anomaly information. An image that does not contain anomaly data and contains only normal data may be labeled for normal information. Specific description of the above-described labeled object is only an example and the present disclosure is not limited thereto.

The object of interest may include anomaly data. The object 301 included in the first image 300 of the first training data may include anomaly data included in the first image 300 of the first training data.

The anomaly data may refer to data that deviates from a normal pattern of data. The data may have an atypical pattern and the anomaly data may refer to data that deviates from this atypical pattern. For example, in a production process, data about an image of a product may have an atypical pattern called a normal product, and anomaly data may be data that deviates from the atypical pattern of a normal product (ie, an image of a bad product, etc.). have. The description of the normal data, the atypical pattern, and the anomaly data of the present disclosure is merely an example and the present disclosure is not limited thereto.

In the above example, the first training data may be training data including a first image 300 for a defective product. The first training data may include a first image 300 that is an input of a neural network model and a label that can be compared with an output of the neural network model. The label of the first training data may be a label based on the object 301 included in the first image 300 input to the neural network model. The label of the first training data may be a label for anomaly data included in the first image 300 input to the neural network model. The detailed description regarding the first learning data described above is merely an example and the present disclosure is not limited thereto.

The processor 120 may identify a label for the anomaly data of the first image 300 included in the first training data.

The processor 120 may identify the object 301 included in the first image 300 included in the first training data. The processor 120 may identify the object 301 included in the first image 300 based on a label for the anomaly data of the first image 300. The processor 120 may perform segmentation on the object 301 based on the label for the anomaly data of the first image 300.

According to an embodiment of the present disclosure, a label matching the first image 300 included in the first training data may be a segmentation label. The processor 120 may perform segmentation on the object 301 by identifying a segmentation label that is a label that matches the first image 300. Segmentation of the object 301 included in the first image 300 may refer to a process of separating a part of an image to be distinguished from another part of the image. The image segmentation may include, for example, separating a portion of the image to be distinguished from other portions based on the edge, color, etc. extracted from the image. In addition, the image segmentation may be a process of identifying the location of the anomaly data from another part of the image by extracting the location information of the anomaly part in the image. In addition, the image segmentation may also include a process of visualizing and displaying a part of the image separated from the other part.

According to another embodiment of the present disclosure, a label matching the first image 300 included in the first training data may be a classification label. When the label matching the first image 300 is a classification label, the processor 120 may receive segmentation information about the object 301 that is the basis of the classification label from another computing device.

For example, data aggregation of data classified as anomaly data among training data including classification labels for each of anomaly data and normal data may be required. The processor 120 may receive or receive segmentation on the anomaly object by transmitting learning data including anomaly data to the user through the network unit 110 or through an interface unit (not shown). .

The processor 120 may transform the first image 300 based on at least a portion of the object 301. The processor 120 may transform the first image 300 to generate a second image. The processor 120 may generate second training data including the second image based on the first training data including the first image 300.

According to an embodiment of the present disclosure, the operation of modifying the first image 300 based on at least a portion of the object 301 by the processor 120 may include the first to include at least a portion of the object 301. And cropping the image 300. The processor 120 may crop the first image 300 to generate second images 310, 311, and 313. The processor 120 may crop the first image 300 to a predetermined size. Can be. The size of the cropped first image 300 may be a predetermined size. The processor 120 may crop the first image 300 to a predetermined size to compute the first image 300 based on the neural network model. The size of the image input to the neural network model may be related to the amount of computation in the network function included in the neural network model. Thus, the processor 120 may crop the first image 300 such that the input of the neural network has the same size. Cropping may mean cutting an area excluding at least a portion of the image to include only at least a portion of the first image 300.

The processor 120 may crop the first image 300 to perform data aggregation of the training data with respect to the anomaly data. When randomly cropping the first image 300 including the anomaly data without other constraints, the first image 300 may be cropped to include only the region excluding the anomaly data, and the second image may not include the anomaly data. It may not. In particular, when performing defect inspection on a semiconductor, for example, the defective portion of the semiconductor can be very small in the image. Therefore, when data aggregation of the semiconductor defect image is performed through general random cropping without any other constraints, the data aggregation is likely to be performed so as not to include the defect portion. For example, if the anomaly region of the semiconductor is very small, the anomaly region is biased to the left of the first image, and cropped to include only the right region of the first image, the second image is anomaly. It may not contain any area at all. When cropping is performed according to an exemplary embodiment of the present disclosure, since cropping is performed so that at least a part of the defective portion is included, efficient data aggregation may be performed.

The processor 120 may generate the second image by modifying the first image 300 to include at least a portion of the object 301. The processor 120 may crop the first image 300 to include at least a portion of the object 301. The processor 120 may crop the first image 300 to include at least a portion of the anomaly data.

The processor 120 may crop the first image 300 such that the first portion 321 included in the object 301 is included in the second images 310, 311, and 313 that are cropped images. The first portion 321 may be at least one area of the object 301 included in the first image 300. The first portion 321 may be a region randomly determined in the object 301. The size of the first portion 321 may be, for example, an area corresponding to one pixel or an area corresponding to larger pixels. For example, the processor 120 may crop the first image 300 such that the second images 310, 311, and 313 include an area corresponding to at least one pixel included in the object 301. . Specific description of the above-mentioned cropping is only an example and the present disclosure is not limited thereto.

The processor 120 crops the first image 300 such that the first portion 321 included in the object 301 overlaps the second portions 323, 325, and 327 of the cropped image. You can ping it. The processor 120 is configured such that the first portion 321 of the first image 300 overlaps the second portions 323, 325, and 327 of the second images 310, 311, and 313, which are cropped images. Image 300 can be cropped. The second images 310, 311, and 313 may be generated based on the cut line shown in FIG. 2A.

2C illustratively shows cropped second images 310, 311, and 313.

The second image 310 is cropped by cropping the first image 300 such that the first portion 321 of the first image 300 overlaps with the second portions 323, 325, and 327 of the second image 310. 311, 313 may be generated.

The size of the first portion 321 of the first image 300 and the size of the second portions 323, 325, 327 of the second image 310 may be the same or may be different. If the size of the second portion 323, 325, 327 of the second image 310 is greater than the size of the first portion 321 of the first image 300, the second portion of the second image 310 ( The first image 300 may be cropped to include the first portion 321 in the areas corresponding to 323, 325, and 327.

2B is an exemplary diagram for explaining a method of determining the second portions 323, 325, and 327 of the second images 310, 311, and 313.

The second portions 323, 325, 327 may be included in the region 320 within a predetermined range at the center of the second image 310, 311, 313, which is the cropped image. The region 320 indicated by a solid line in the second images 310, 311, and 313 illustrated in FIG. 2B may be an area except a buffer region where the second portions 323, 325, and 327 should not be located. The second portions 323, 325, and 327 may be included in regions other than the buffer regions of the second images 310, 311, and 313 that are cropped images. For example, the buffer area may be an area corresponding to 20 percent area from the outer edges of the second images 310, 311, and 313. In FIG. 2B, the region 320 in which the second portions 323, 325, and 327 are to be included is illustrated in a quadrangular shape. However, the present invention is not limited thereto and may have other shapes. Specific description of the above-described buffer region is only an example and the present disclosure is not limited thereto.

When the second portions 323, 325, and 327 are determined without the buffer area, only a portion of the anomaly data may be included in the second images 310, 311, and 313. If only a small portion of anomaly data is included, it can be difficult to use as anomaly training data. For example, when the first portion 321 is determined as the leftmost region of the object 301 included in the first image 300, and the second portion is determined as the rightmost region of the second image 310, Only a portion of the anomaly data may be included in the second image 310. Accordingly, the present disclosure may determine the second portions 323, 325, and 327 in consideration of the buffer area so that the second image 310 includes meaningful anomaly data that can be used as the anomaly learning data.

The first image 300 is cropped to overlap the first portion 220 and the second portions 323, 325, and 327 to generate a second image 310 included in the second training data, which is new training data. Can be.

According to an embodiment of the present disclosure, the operation of the processor 120 modifying the image based on at least a part of the object may include the operation of the processor 120 modifying a portion of the image except at least a portion of the object. Can be. A description with reference to FIG. 3 is as follows. 3 is a diagram illustrating an example of a method of generating training data according to an embodiment of the present disclosure.

The processor 120 may mask a portion of the object 301 except at least a portion thereof. Masking may be an area designated to be excluded from the calculation when the neural network model is calculated. The processor 120 may generate second training data by generating a third image 400 masking a portion except at least a portion of the object 301. When the third image 400 is calculated using the neural network model, the processor 120 may calculate the portion except the masked portion of the second image. In FIG. 3, the masked portion is a portion marked in black in the image 400. For example, as shown in FIG. 3, the processor 120 may mask all portions except the object 301. For example, the processor 120 may mask at least one portion except for the object 301. For example, the processor 120 may mask at least some of the portions other than the first portion 321 included in the object 301. Specific description of the above-mentioned masking is only an example and the present disclosure is not limited thereto.

The processor 120 may generate second training data including the modified image. The processor 120 may generate the second training data by generating an image cropped to include at least a part of the object of interest based on the first training data. The processor 120 may generate second training data by generating an image masked to exclude at least a portion of the object of interest based on the first training data.

The processor 120 may generate a training data set including the first training data and the second training data. The processor 120 may perform data aggregation based on the first training data included in the existing training data set. The processor 120 may generate a new training data set including first training data included in the existing training data set and second training data generated through data aggregation.

The processor 120 may train a model including one or more network functions to identify an object included in an image input to the model based on the training data set.

According to one embodiment of the present disclosure, the training data set may include first training data and second training data. Each of the at least two training data included in the training data set may include a defect inspection target product image and information labeled whether a defect is based on the product image. For example, when the second training data is generated based on the first image and the label included in the first training data, the second training data may include the second image and the label modified from the first image.

The processor 120 inputs the training data to an input node of the neural network model to learn and generate the neural network model, and based on the defect presence data calculated at the output layer of the neural network model and the label included in the training data. The weight of the neural network model can be adjusted. The processor 120 propagates from the output layer included in the one or more network functions of the neural network model through the one or more hidden layers to the input layer based on the error, thereby updating the weights set in each link to train the neural network model. Can be.

4 is a flowchart of a method of generating training data according to an embodiment of the present disclosure.

The computing device 100 may perform data aggregation based on the first training data for the class lacking the training data. The computing device 100 may identify an object included in the image based on the image included in the first training data (610).

The computing device 100 may identify the object of interest included in the image.

The object of interest may include a labeled object.

The label according to an embodiment of the present disclosure may be a segmentation label. The segmentation label may indicate a partial region included in the image to be distinguished from other regions of the image.

The label according to another embodiment of the present disclosure may be a classification label. The classification label may indicate information about which class the image belongs to. The classification label may indicate whether the image includes specific information that the user wants to acquire from the image.

The object of interest may include anomaly data. The anomaly data may refer to anomaly data that deviates from a normal pattern of data. The data may have an atypical pattern and the anomaly data may refer to data that deviates from this atypical pattern.

The computing device 100 may perform segmentation on objects included in the image.

The computing device 100 may transform 620 the image based on at least a portion of the object.

Computing device 100 may crop the image to include at least a portion of an object. The computing device 100 may crop the image such that the first portion included in the object is included in the cropped image. The computing device 100 may crop the image such that the first portion included in the object and the second portion of the cropped image overlap.

The second portion may be a portion included in an area within a predetermined range at the center of the cropped image.

The computing device 100 may modify portions of the image except at least some of the objects. The computing device 100 may mask a portion except at least a portion of the object.

The computing device 100 may generate second training data including the modified image (630).

The computing device 100 may generate a training data set including the first training data and the second training data.

The computing device 100 may train a model including one or more network functions to identify an object included in an image input to the model based on the training data set.

5 is a block diagram of a computing device according to an embodiment of the present disclosure.

5 shows a brief general schematic diagram of an example computing environment in which embodiments of the present disclosure may be implemented.

Although the present disclosure has been described above generally with respect to computer executable instructions that may be executed on one or more computers, those skilled in the art will appreciate that the present disclosure may be implemented in combination with other program modules and / or as a combination of hardware and software. will be.

Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will appreciate that the methods of the present disclosure may include uniprocessor or multiprocessor computer systems, minicomputers, mainframe computers, as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like (each of which And other computer system configurations, including one or more associated devices.

The described embodiments of the present disclosure can also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

Computers typically include a variety of computer readable media. Any medium accessible by a computer can be a computer readable medium. Computer readable media includes volatile and nonvolatile media, transitory and non-transitory media, removable and non-removable media. By way of example, and not limitation, computer readable media may comprise computer readable storage media and computer readable transmission media. Computer-readable storage media are volatile and nonvolatile media, temporary and non-transitory media, removable and non-removable implemented in any method or technology for storing information such as computer readable instructions, data structures, program modules or other data. Media. Computer storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROMs, digital video disks or other optical disk storage devices, magnetic cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, Or any other medium that can be accessed by a computer and used to store desired information.

Computer-readable transmission media typically embody computer readable instructions, data structures, program modules or other data in a modulated data signal, such as other transport mechanisms, and include all information delivery media. do. The term modulated data signal refers to a signal that has one or more of its characteristics set or changed to encode information in the signal. By way of example, and not limitation, computer readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, or other wireless media. Combinations of any of the above should also be included within the scope of computer readable transmission media.

An exemplary environment 1100 is illustrated that implements various aspects of the present disclosure, including a computer 1102, which includes a processing unit 1104, a system memory 1106, and a system bus 1108. do. System bus 1108 connects system components, including but not limited to system memory 1106, to processing unit 1104. Processing unit 1104 may be any of a variety of commercially available processors. Dual processor and other multiprocessor architectures may also be used as the processing unit 1104.

System bus 1108 may be any of several types of bus structures that may be further interconnected to a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 1106 includes read only memory (ROM) 1110 and random access memory (RAM) 1112. The basic input / output system (BIOS) is stored in nonvolatile memory 1110, such as ROM, EPROM, EEPROM, etc., and the BIOS provides a basic aid for transferring information between components in the computer 1102, such as during startup. Contains routines. RAM 1112 may also include fast RAM, such as static RAM, for caching data.

Computer 1102 also includes an internal hard disk drive (HDD) 1114 (eg, EIDE, SATA) —the internal hard disk drive 1114 may also be configured for external use within a suitable chassis (not shown). Yes, magnetic floppy disk drive (FDD) 1116 (eg, for reading from or writing to removable diskette 1118), and optical disk drive 1120 (eg, CD-ROM Disc 1122 for reading from or writing to or writing from other high capacity optical media such as DVD). The hard disk drive 1114, the magnetic disk drive 1116, and the optical disk drive 1120 are connected to the system bus 1108 by the hard disk drive interface 1124, the magnetic disk drive interface 1126, and the optical drive interface 1128, respectively. ) Can be connected. Interface 1124 for external drive implementation includes at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies.

These drives and their associated computer readable media provide nonvolatile storage of data, data structures, computer executable instructions, and the like. In the case of computer 1102, drives and media correspond to storing any data in a suitable digital format. Although the description of computer readable media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art will appreciate zip drives, magnetic cassettes, flash memory cards, cartridges, and the like. Other types of media readable by the computer, etc. may also be used in the exemplary operating environment and it will be appreciated that any such media may include computer executable instructions for performing the methods of the present disclosure.

Multiple program modules may be stored in the drive and RAM 1112, including operating system 1130, one or more application programs 1132, other program modules 1134, and program data 1136. All or a portion of the operating system, applications, modules and / or data may also be cached in RAM 1112. It will be appreciated that the present disclosure may be implemented in various commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 1102 through one or more wired / wireless input devices, such as a keyboard 1138 and a mouse 1140. Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. These and other input devices are often connected to the processing unit 1104 through an input device interface 1142 that is connected to the system bus 1108, but the parallel port, IEEE 1394 serial port, game port, USB port, IR interface, Etc. can be connected by other interfaces.

A monitor 1144 or other type of display device is also connected to the system bus 1108 via an interface such as a video adapter 1146. In addition to the monitor 1144, the computer generally includes other peripheral output devices (not shown) such as speakers, printers, and the like.

Computer 1102 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer (s) 1148, via wired and / or wireless communications. Remote computer (s) 1148 can be a workstation, computing device computer, router, personal computer, portable computer, microprocessor-based entertainment device, peer device, or other conventional network node, and typically is associated with computer 1102. Although many or all of the components described above are included, for simplicity, only memory storage 1150 is shown. The logical connections shown include wired / wireless connections to a local area network (LAN) 1152 and / or a larger network, such as a telecommunications network (WAN) 1154. Such LAN and WAN networking environments are commonplace in offices and businesses, facilitating enterprise-wide computer networks such as intranets, all of which may be connected to worldwide computer networks, such as the Internet.

When used in a LAN networking environment, the computer 1102 is connected to the local network 1152 via a wired and / or wireless communication network interface or adapter 1156. Adapter 1156 may facilitate wired or wireless communication to LAN 1152, which also includes a wireless access point installed therein for communicating with wireless adapter 1156. When used in a WAN networking environment, the computer 1102 may include a modem 1158, connect to a communications computing device on the WAN 1154, or establish communications over the WAN 1154, such as over the Internet. Other means. The modem 1158, which may be an internal or external and wired or wireless device, is connected to the system bus 1108 via the serial port interface 1142. In a networked environment, program modules or portions thereof described with respect to computer 1102 may be stored in remote memory / storage device 1150. It will be appreciated that the network connections shown are exemplary and other means of establishing a communications link between the computers may be used.

Computer 1102 is associated with any wireless device or entity disposed and operating in wireless communication, such as a printer, scanner, desktop and / or portable computer, portable data assistant, communications satellite, wireless detectable tag. Communicate with any equipment or location and telephone. This includes at least Wi-Fi and Bluetooth wireless technology. Thus, the communication can be a predefined structure as in a conventional network or simply an ad hoc communication between at least two devices.

Wireless Fidelity (Wi-Fi) allows you to connect to the Internet without wires. Wi-Fi is a wireless technology such as a cell phone that allows a device, for example, a computer, to transmit and receive data indoors and outdoors, ie anywhere within the coverage area of a base station. Wi-Fi networks use a wireless technology called IEEE 802.11 (a, b, g, etc.) to provide secure, reliable, high-speed wireless connections. Wi-Fi may be used to connect computers to each other, to the Internet, and to a wired network (using IEEE 802.3 or Ethernet). Wi-Fi networks can operate in unlicensed 2.4 and 5 GHz wireless bands, for example, at 11 Mbps (802.11a) or 54 Mbps (802.11b) data rates, or in products that include both bands (dual band). have.

Those skilled in the art will appreciate that information and signals may be represented using any of a variety of different technologies and techniques. For example, data, instructions, instructions, information, signals, bits, symbols, and chips that may be referenced in the above description may include voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields. Or particles, or any combination thereof.

One of ordinary skill in the art of the disclosure will appreciate that the various illustrative logical blocks, modules, processors, means, circuits and algorithm steps described in connection with the embodiments disclosed herein may be implemented in electronic hardware, It will be appreciated that for purposes of the present invention, various forms of program or design code, or combinations thereof, may be implemented. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the particular application and design constraints imposed on the overall system. One skilled in the art of the present disclosure may implement the described functionality in various ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present disclosure.

The various embodiments presented herein may be embodied in a method, apparatus, or article of manufacture using standard programming and / or engineering techniques. The term "article of manufacture" includes a computer program or media accessible from any computer-readable device. For example, computer-readable media may include magnetic storage devices (eg, hard disks, floppy disks, magnetic strips, etc.), optical discs (eg, CDs, DVDs, etc.), smart cards, and flash memory. Devices, such as, but not limited to, EEPROM, cards, sticks, key drives, and the like. In addition, various storage media presented herein include one or more devices and / or other machine-readable media for storing information.

It is to be understood that the specific order or hierarchy of steps in the processes presented is an example of exemplary approaches. Based upon design priorities, it is understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The accompanying method claims present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

The description of the presented embodiments is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these embodiments will be apparent to those of ordinary skill in the art. The general principles defined herein may be applied to other embodiments without departing from the scope of the present disclosure. Thus, the present disclosure should not be limited to the embodiments set forth herein but should be construed in the broadest scope consistent with the principles and novel features set forth herein.

Claims (14)

A computer program stored in a computer readable storage medium, the computer program configured to perform operations for providing a method of generating learning data when executed on one or more processors of a computing device, the operations comprising:
Identifying an object included in the image based on the image included in the first training data;
Cropping the image such that at least a portion of the second portion included in an area within a predetermined range of the cropped image overlaps with the first portion of the object; And
Generating second training data comprising the cropped image;
Including,
Computer program stored on a computer readable storage medium.
The method of claim 1,
Identifying an object included in the image based on the image included in the first training data,
Identifying an object of interest included in the image;
Including,
Computer program stored on a computer readable storage medium.
The method of claim 2,
The object of interest comprises a labeled object,
Computer program stored on a computer readable storage medium.
The method of claim 2,
The object of interest includes anomaly data,
Computer program stored on a computer readable storage medium.
The method of claim 1,
Identifying an object included in the image based on the image included in the first training data,
Performing segmentation on an object included in the image;
Including,
Computer program stored on a computer readable storage medium.
delete delete The method of claim 1,
Cropping the image such that at least a portion of the second portion included in an area within a predetermined range that is at least a portion of the cropped image and the first portion of the object overlaps,
Cropping the image such that a first portion included in the object and a second portion of the cropped image overlap;
Including,
Computer program stored on a computer readable storage medium.
The method of claim 1,
The second portion is included in an area within a predetermined range at the center of the cropped image,
Computer program stored on a computer readable storage medium.
The method of claim 1,
Modifying a portion of the image except at least a portion of the object; And
Generating second training data including the modified image;
Including more;
Computer program stored on a computer readable storage medium.
The method of claim 1,
Masking a portion of the object other than at least a portion; And
Generating second training data including the image modified by the masking;
Including more;
Computer program stored on a computer readable storage medium.
The method of claim 1,
Generating a training data set comprising the first training data and the second training data; And
Training a model including at least one network function to identify an object included in an image input to the model based on the training data set;
Including more;
Computer program stored on a computer readable storage medium.
As a training data generation method,
Identifying, by a processor included in a computing device, an object included in the image based on the image included in the first training data;
The processor cropping the image such that at least a portion of the second portion included in an area within a predetermined range of the cropped image overlaps with the first portion of the object; And
Generating, by the processor, second training data including the cropped image;
Including,
How to generate training data.
As a server for providing a training data generation method,
A processor including one or more cores; And
Memory;
Including,
The processor,
Identify an object included in the image based on the image included in the first training data,
Cropping the image such that at least a portion of the second portion included in an area within a predetermined range of the cropped image overlaps with the first portion of the object, and
Generating second training data comprising the cropped image,
Server for providing training data generation method.

Images