JP7266674B2 - Image classification model training method, image processing method and apparatus - Google Patents

Description

(Cross reference to related applications)
This application is filed with the Patent Office of China on Nov. 11, 2019, with application number CN201911097069. X and entitled "Method for Training Image Classification Model, Image Processing Method and Apparatus", the entire content of which is incorporated herein by reference.

TECHNICAL FIELD The present application relates to the field of data processing technology, and in particular to an image classification model training method, image processing method and apparatus.

Before performing a classification task based on a neural network model, we first need to train the neural network model. When training neural network models, we generally need to rely on a large number of labeled samples to improve the classification accuracy of the neural network model.

In the process of labeling samples, the number of positive samples corresponding to different categories may be uneven.

Embodiments of the present application provide at least an image classification model training method, image processing method and apparatus.

According to a first aspect, embodiments of the present application provide a method for training an image classification model. The method is obtaining a set of labeled sample images, wherein the set of labeled sample images includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image; determining frequency information respectively corresponding to each of a plurality of predetermined categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images; and the plurality of labeled sample images. determining loss information for an image classification model based on the classification labels respectively corresponding to and the frequency information respectively corresponding to each category; and training the image classification model based on the loss information. , wherein the image classification model is for determining categories of images to be recognized.

Here, when training the classification model, based on the labeled sample images and the classification labels respectively corresponding to each labeled sample image, frequency information corresponding to each of the plurality of predetermined categories is determined. Then, based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category, loss information of the classification model is determined, and the classification model is trained based on the loss information. In the process, image classification is performed in the training process of an image classification model based on the frequency information of each category to determine the loss information according to the frequency information corresponding to each category and the classification label respectively corresponding to each sample image. You can adjust the degree of influence of positive and negative samples for a particular category on the model, which balances the influence of positive and negative samples on each category, and further refines the classification model for categories with few positive samples. can improve the recognition accuracy of

In an optional embodiment, determining frequency information corresponding to each of a predetermined plurality of categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images comprises: for each category, based on the labeled sample images belonging to the category and the original sample images corresponding to each labeled sample image, determining the number of original sample images corresponding to the category; sample images corresponding to one or more labeled sample images; and determining frequency information corresponding to the category based on the number of original sample images. Here, frequency information is determined by such a method to avoid determining a category with few positive samples as a background category.

In an optional embodiment, determining frequency information corresponding to each of a predetermined plurality of categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images comprises: Determining, for each category, frequency information corresponding to the category based on the total number of labeled sample images in the labeled sample image set and the number of labeled sample images belonging to the category, or as frequency information corresponding to the category. Here, based on the number of labeled sample images, determine the frequency information corresponding to the category, the process is simpler and faster.

In an optional embodiment, determining loss information for a classification model based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category includes: , obtaining loss information when classifying the labeled sample images using a classification model based on the classification labels of the labeled sample images and frequency information corresponding to each category, respectively . Now, the loss information is determined for each labeled sample image. It realizes adjusting the parameters of the image classification model based on each labeled sample image, and finally realizes the training of the image classification model.

In an optional embodiment, obtaining loss information when classifying the labeled sample images using a classification model based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category. Determining a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample image based on the classification label of the labeled sample image and frequency information corresponding to each category. and obtaining loss information when classifying the labeled sample image using the classification model based on a loss weight corresponding to each category of a plurality of predetermined categories. include. Here, based on the classification label and frequency information, loss information is determined to improve the recognition accuracy of the image classification model for categories with few positive samples.

In an optional embodiment, for the labeled sample images, based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category, Determining a loss weight includes, for any one category, the classification label of the labeled sample image indicating that the labeled sample image is a non-background labeled sample image; Determining a loss weight corresponding to the one category as a first predetermined loss weight value if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold. Here, during the training process of the image classification model, the degree of influence of the positive and negative samples of a specific category on the image classification model can be adjusted, thereby improving the detection accuracy of the classification model for categories with few positive samples. Let

In an optional embodiment, for the labeled sample images, based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category, Determining a loss weight includes, for any one category, the classification label of the labeled sample image indicating that the labeled sample image is a background labeled sample image, and/or Determining a loss weight corresponding to the one category as a second predetermined loss weight value if it is determined that frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold. Here, during the training process of the image classification model, the degree of influence of the positive and negative samples of a specific category on the image classification model can be adjusted, thereby improving the detection accuracy of the classification model for categories with few positive samples. Let

In an optional embodiment, for the labeled sample images, based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category, Determining a loss weight is based on classification labels of the labeled sample images, frequency information respectively corresponding to each category, positive sample category sets and negative sample category sets of original sample images corresponding to the labeled sample images. , determining a loss weight corresponding to each category of a predetermined plurality of categories for the labeled sample image.

wherein the positive sample category set includes at least one target target category included in the original sample image, and the negative sample category set includes at least one target target category not included in the original sample image is included. Now, it is determined whether any one category belongs to the positive sample set and the negative sample category set corresponding to the labeled sample image. Doing so does not reduce or ignore the impact of labeled sample images on that category, and also allows the image classification model to learn more and differential features of any one category. and improve the accuracy of the image classification model.

In an optional embodiment, determining from the set of labeled sample images a target labeled sample image corresponding to the same original sample image as the labeled sample image, and combining the labeled sample image and the target labeled sample image. a method for determining a positive sample category set of the original sample image corresponding to the labeled sample image based on the classification labels respectively corresponding to for determining a positive sample category set of an original sample image corresponding to the labeled sample image based on: Obtain the positive sample category set of the original sample image corresponding to the labeled sample image by using the method of . We now easily and quickly determine the positive sample category set corresponding to each labeled sample image.

In an optional embodiment, using a scheme for determining a negative sample category set of an original sample image corresponding to the labeled sample image based on a second auxiliary classification label of the original sample image corresponding to the labeled sample image. to obtain a negative sample category set of the original sample image corresponding to the labeled sample image, wherein the second sub-classification label is for indicating the target object category not included in the original sample image. We now easily and quickly determine the negative sample category set corresponding to each labeled sample image.

In an optional embodiment, based on the classification labels of the labeled sample images, frequency information respectively corresponding to each category, positive sample category sets and negative sample category sets of original sample images corresponding to the labeled sample images, Determining, for the labeled sample image, a loss weight corresponding to each category of a plurality of predetermined categories includes, for any one category, the classification label of the labeled sample image being the indicating that the labeled sample image is a non-background labeled sample image, the one category does not belong to the positive sample category set and the negative sample category set, and corresponds to the one category Determining a loss weight corresponding to the one of the categories as a first predetermined loss weight value if the frequency information is determined to be less than a predetermined frequency threshold.

In an optional embodiment, based on the classification labels of the labeled sample images, frequency information respectively corresponding to each category, positive sample category sets and negative sample category sets of original sample images corresponding to the labeled sample images, Determining, for the labeled sample image, a loss weight corresponding to each category of a plurality of predetermined categories includes, for any one category, the classification label of the labeled sample image being the indicating that the labeled sample images are background labeled sample images; and/or the one category belongs to the positive sample category set or the negative sample category set; and/or any one determining a loss weight corresponding to the one category as a second predetermined loss weight value if it is determined that the frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold. Now, it is determined whether any one category belongs to the positive sample category set and the negative sample category set corresponding to the labeled sample image. Doing so does not reduce or ignore the impact of labeled sample images on that category, and also allows the image classification model to learn more and differential features of any one category. and improve the accuracy of the image classification model.

According to a second aspect, embodiments of the present application provide an image processing method. The method comprises performing an image processing task based on a pre-trained image classification model, said image classification model being a method of training an image classification model according to any one of the first aspects above. trained, the image processing task includes one or more of image classification, object detection, keypoint detection, image segmentation, instance segmentation.

According to a third aspect, embodiments of the present application provide a method for training a data classification model. The method is obtaining a labeled sample data set, wherein the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image; determining frequency information respectively corresponding to each of a plurality of predetermined categories based on labeled sample data and classification labels respectively corresponding to each of the labeled sample data; determining loss information for a data classification model based on the classification labels respectively corresponding to and the frequency information respectively corresponding to each category; and training the data classification model based on the loss information. , wherein the data classification model is for determining categories of data to be recognized.

In an optional embodiment, determining frequency information corresponding to each of a predetermined plurality of categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data comprises: For each category, based on the labeled sample data belonging to the category and the original sample data corresponding to each labeled sample data, determining the number of original sample data corresponding to the category, sample data corresponding to one or more labeled sample data; and determining frequency information corresponding to the category based on the number of original sample data.

In an optional embodiment, determining frequency information corresponding to each of a predetermined plurality of categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data comprises: Determining, for each category, frequency information corresponding to the category based on the total number of labeled sample data in the labeled sample data set and the number of labeled sample data belonging to the category, or as frequency information corresponding to the category.

In an optional embodiment, determining loss information for a classification model based on the classification labels respectively corresponding to the plurality of labeled sample data and the frequency information respectively corresponding to each category comprises: , obtaining loss information when classifying the labeled sample data using a classification model based on the classification label of the labeled sample data and the frequency information corresponding to each category, respectively .

In an optional embodiment, obtaining loss information when classifying the labeled sample data using a classification model based on the classification labels of the labeled sample data and frequency information corresponding to each category, respectively. Determining a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample data based on the classification label of the labeled sample data and frequency information corresponding to each category. and obtaining loss information when classifying the labeled sample data using the classification model based on a loss weight corresponding to each category of a plurality of predetermined categories. include.

In an optional embodiment, for the labeled sample data, based on the classification label of the labeled sample data and the frequency information corresponding to each category, Determining a loss weight includes, for any one category, the classification label of the labeled sample data indicating that the labeled sample data is non-background labeled sample data; Determining a loss weight corresponding to the one category as a first predetermined loss weight value if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold.

In an optional embodiment, for the labeled sample data, based on the classification label of the labeled sample data and the frequency information corresponding to each category, Determining a loss weight includes, for any one category, the classification label of the labeled sample data indicating that the labeled sample data is a background labeled sample image; Determining a loss weight corresponding to the one category as a second predetermined loss weight value if it is determined that frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold.

In an optional embodiment, for the labeled sample data, based on the classification label of the labeled sample data and the frequency information corresponding to each category, Determining a loss weight is based on classification labels of the labeled sample data, frequency information respectively corresponding to each category, positive sample category sets and negative sample category sets of original sample data corresponding to the labeled sample data. , determining a loss weight corresponding to each of a predetermined plurality of categories for the labeled sample data.

wherein the positive sample category set includes at least one target category included in the original sample data, and the negative sample category set includes at least one target category not included in the original sample data is included.

In an optional embodiment, determining target labeled sample data corresponding to the same original sample data as the labeled sample data from the set of labeled sample data, and determining the labeled sample data and the target labeled sample data. A method for determining a positive sample category set of the original sample data corresponding to the labeled sample data based on the classification labels corresponding to each, or a first auxiliary classification label of the original sample data corresponding to the labeled sample data a positive sample category set of the original sample data corresponding to the labeled sample data based on: is used to obtain the positive sample category set of the original sample data corresponding to the labeled sample data.

In an optional embodiment, using a method of determining a negative sample category set of original sample data corresponding to the labeled sample data based on a second auxiliary classification label of the original sample data corresponding to the labeled sample data. to obtain a negative sample category set of the original sample data corresponding to the labeled sample data, and the second sub-classification label is for indicating the category of the target object that is not included in the original sample data.

In an optional embodiment, based on the classification label of the labeled sample data, the frequency information corresponding to each category, the positive sample category set and the negative sample category set of the original sample data corresponding to the labeled sample data, Determining a loss weight corresponding to each category of a plurality of predetermined categories for the labeled sample data includes determining, for any one category, the classification label of the labeled sample data as the Indicates that the labeled sample data is non-background labeled sample data, the one category does not belong to the positive sample category set and the negative sample category set, and corresponds to the one category Determining a loss weight corresponding to the one of the categories as a first predetermined loss weight value if the frequency information is determined to be less than a predetermined frequency threshold.

In an optional embodiment, based on the classification label of the labeled sample data, the frequency information corresponding to each category, the positive sample category set and the negative sample category set of the original sample data corresponding to the labeled sample data, Determining a loss weight corresponding to each category of a plurality of predetermined categories for the labeled sample data includes determining, for any one category, the classification label of the labeled sample data as the indicates that the labeled sample data is background labeled sample data, and/or any one category belongs to the positive sample category set or the negative sample category set, and/or any one determining a loss weight corresponding to the one category as a second predetermined loss weight value if it is determined that the frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold.

According to a fourth aspect, embodiments of the present application provide a data processing method. The method comprises performing a data processing task based on a pre-trained data classification model, said data classification model being a method of training a data classification model according to any one of the second aspects above. trained, the data processing task includes one or more of data classification, data segmentation, instance segmentation, and the data includes any one of image data and text data.

According to a fifth aspect, embodiments of the present application provide an image classification model training apparatus. The apparatus includes a first acquisition module configured to acquire a set of labeled sample images, wherein the set of labeled sample images includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image. and determining frequency information respectively corresponding to each category of a predetermined plurality of categories based on the labeled sample images and classification labels respectively corresponding to each labeled sample image. and determining loss information of an image classification model based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category. and a first training module configured to train the image classification model based on the loss information, wherein the image classification model is to be recognized a first training module for determining categories of images.

In an optional embodiment, the first frequency information determination module corresponds to each of a predetermined plurality of categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images. When determining the frequency information for each category, determine the number of original sample images corresponding to the category based on the labeled sample images belonging to the category and the original sample images corresponding to each labeled sample image wherein each original sample image corresponds to one or more labeled sample images; and determining frequency information corresponding to the category based on the number of the original sample images. , is configured to run

In an optional embodiment, the first frequency information determination module corresponds to each of a predetermined plurality of categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images. frequency information corresponding to each category based on the total number of labeled sample images in the labeled sample image set and the number of labeled sample images belonging to the category; Alternatively, the number of labeled sample images belonging to the category is used as frequency information corresponding to the category.

In an optional embodiment, the first loss information determination module determines loss information for a classification model based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category. When classifying each labeled sample image using a classification model based on the frequency information corresponding to the classification label of the labeled sample image and each category, respectively is configured to obtain loss information for

In an optional embodiment, the first loss information determination module uses a classification model for the labeled sample images based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category. When obtaining loss information when performing classification, based on the classification label of the labeled sample image and the frequency information corresponding to each category, for the labeled sample image, each of a plurality of predetermined categories Loss information when a loss weight corresponding to a category is determined, and the labeled sample image is classified using the classification model based on the loss weight corresponding to each category of a plurality of predetermined categories. is configured to obtain

In an optional embodiment, the first loss information determination module performs a plurality of predetermined loss information determination on the labeled sample images based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category. When determining the loss weight corresponding to each of the categories, for any one category the classification label of the labeled sample image is the labeled sample image with no background and determining that the frequency information corresponding to the one category is less than the predetermined frequency threshold, then determining the loss weight corresponding to the one category as a first predetermined loss weight value. configured to

In an optional embodiment, the first loss information determination module performs a plurality of predetermined loss information determination on the labeled sample images based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category. When determining the loss weight corresponding to each of the categories, the classification label of the labeled sample image indicates that the labeled sample image is a background labeled sample image, and/or any or determining that the frequency information corresponding to one category is equal to or greater than a predetermined frequency threshold, the loss weight corresponding to the one category is determined as a second predetermined loss weight value.

In an optional embodiment, the first loss information determination module performs a plurality of predetermined loss information determination on the labeled sample images based on the classification labels of the labeled sample images and frequency information respectively corresponding to each category. When determining the loss weight corresponding to each category of categories, the classification label of the labeled sample image, the frequency information corresponding to each category, and the positive sample category set of the original sample image corresponding to the labeled sample image. and a loss weight corresponding to each category of a predetermined plurality of categories for the labeled sample image based on the set of negative sample categories.

wherein the positive sample category set includes at least one target target category included in the original sample image, and the negative sample category set includes at least one target target category not included in the original sample image is included.

In an optional embodiment, the first loss information determination module determines from the set of labeled sample images a target labeled sample image corresponding to the same original sample image as the labeled sample image; A method for determining a positive sample category set of an original sample image corresponding to the labeled sample image or corresponding to the labeled sample image, based on the classification labels respectively corresponding to the sample image and the target labeled sample image; A method for determining a positive sample category set of an original sample image corresponding to the labeled sample image based on a first auxiliary classification label of the original sample image, wherein the first auxiliary classification label is associated with the original sample image. It is configured to obtain a positive sample category set of original sample images corresponding to the labeled sample images using a scheme that is for indicating the category of the target object included.

In an optional embodiment, the first loss information determination module determines the negative of the original sample image corresponding to the labeled sample image based on a second auxiliary classification label of the original sample image corresponding to the labeled sample image. obtaining a negative sample category set of an original sample image corresponding to the labeled sample image, using a method for determining a sample category set, wherein the second auxiliary classification label is included in the original sample image; It is intended to indicate a category of targets that are not targeted.

In an optional embodiment, the first loss information determination module includes classification labels of the labeled sample images, frequency information respectively corresponding to each category, and a positive sample category set of original sample images corresponding to the labeled sample images. and determining a loss weight corresponding to each of a predetermined plurality of categories for the labeled sample image based on the set of negative sample categories;
For any one category, the classification label of the labeled sample image indicates that the labeled sample image is a non-background labeled sample image, and the any one category is the positive sample category set and when it is determined that the frequency information corresponding to any one category does not belong to the negative sample category set and is less than a predetermined frequency threshold, a loss weight corresponding to the one category is set to a first predetermined is configured to be determined as the loss weight value of

In an optional embodiment, the first loss information determination module includes classification labels of the labeled sample images, frequency information respectively corresponding to each category, and a positive sample category set of original sample images corresponding to the labeled sample images. and based on the negative sample category set, for the labeled sample image, when determining the loss weight corresponding to each category of a plurality of predetermined categories, for any one category, the labeled the classification label of the sample image indicates that the labeled sample image is a background labeled sample image, and/or the any one category belongs to the positive sample category set or the negative sample category set; and/or determining the loss weight corresponding to the one category as a second predetermined loss weight value when it is determined that the frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold. configured as

According to a sixth aspect, the present application provides an image processing apparatus. The apparatus comprises a first processing module configured to perform an image processing task based on a pretrained image classification model, the image classification model according to any one of the first aspects above. wherein the image processing task includes one or more of image classification, object detection, keypoint detection, image segmentation, instance segmentation.

According to a seventh aspect, embodiments of the present application provide a data classification model training apparatus. The apparatus includes a second acquisition module configured to acquire a labeled sample data set, wherein the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image. and determining frequency information respectively corresponding to each category of a predetermined plurality of categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data. and determining loss information of a data classification model based on the classification labels respectively corresponding to the plurality of labeled sample data and the frequency information respectively corresponding to each category. and a second training module configured to train the data classification model based on the loss information, wherein the data classification model is to be recognized a second training module for determining categories of data.

In an optional embodiment, the second frequency information determination module corresponds to each of a predetermined plurality of categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data. When determining frequency information for each category, determine the number of original sample data corresponding to the category based on the labeled sample data belonging to the category and the original sample data corresponding to each labeled sample data wherein each original sample data corresponds to one or more labeled sample data; and based on the number of the original sample data, determining frequency information corresponding to the category. , is configured to run

In an optional embodiment, the second frequency information determination module corresponds to each of a predetermined plurality of categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data. When determining frequency information for each category, based on the total number of labeled sample data in the labeled sample data set and the number of labeled sample data belonging to the category, frequency information corresponding to the category is determined. Alternatively, the number of labeled sample data belonging to the category is used as frequency information corresponding to the category.

In an optional embodiment, the second loss information determination module determines loss information for a classification model based on the classification labels respectively corresponding to the plurality of labeled sample data and the frequency information respectively corresponding to each category. In the case where each labeled sample data is classified using a classification model based on the frequency information corresponding to the classification label of the labeled sample data and each category. is configured to obtain loss information for

In an optional embodiment, the second loss information determination module uses a classification model for the labeled sample data based on the classification labels of the labeled sample data and frequency information respectively corresponding to each category. When obtaining loss information when performing classification, based on the classification label of the labeled sample data and the frequency information corresponding to each category, for the labeled sample data, each of a plurality of predetermined categories Loss information when a loss weight corresponding to a category is determined, and the labeled sample data is classified using the classification model based on the loss weight corresponding to each category of a plurality of predetermined categories. is configured to obtain

In an optional embodiment, the second loss information determination module performs a predetermined plurality of operations on the labeled sample data based on the classification label of the labeled sample data and frequency information respectively corresponding to each category. When determining the loss weight corresponding to each category of categories, for any one category the classification label of the labeled sample data is the non-background labeled sample data and determining that the frequency information corresponding to the one category is less than the predetermined frequency threshold, then determining the loss weight corresponding to the one category as a first predetermined loss weight value. configured to

In an optional embodiment, the second loss information determination module performs a predetermined plurality of operations on the labeled sample data based on the classification label of the labeled sample data and frequency information respectively corresponding to each category. When determining the loss weight corresponding to each of the categories, for any one category, the classification label of the labeled sample data is the labeled sample image of the background. and/or if it is determined that the frequency information corresponding to the one category is equal to or greater than a predetermined frequency threshold, the loss weight corresponding to the one category is set to a second predetermined loss weight value is configured to determine as

In an optional embodiment, the second loss information determination module performs a predetermined plurality of operations on the labeled sample data based on the classification label of the labeled sample data and frequency information respectively corresponding to each category. When determining the loss weight corresponding to each category of the categories, the classification label of the labeled sample data, the frequency information corresponding to each category, and the positive sample category set of the original sample data corresponding to the labeled sample data and a loss weight corresponding to each category of a predetermined plurality of categories for the labeled sample data based on the set of negative sample categories.

wherein the positive sample category set includes at least one target category included in the original sample data, and the negative sample category set includes at least one target category not included in the original sample data is included.

In an optional embodiment, the second loss information determination module determines target labeled sample data corresponding to the same original sample data as the labeled sample data from the labeled sample data set; A method for determining a positive sample category set of original sample data corresponding to the labeled sample data based on the classification labels respectively corresponding to the sample data and the target labeled sample data, or corresponding to the labeled sample data. A method for determining a positive sample category set of original sample data corresponding to the labeled sample data based on a first auxiliary classification label of the original sample data, wherein the first auxiliary classification label is associated with the original sample data It is configured to obtain a positive sample category set of original sample data corresponding to the labeled sample data using a scheme that is for indicating the category of the target subject included.

In an optional embodiment, the second loss information determination module determines the negative of the original sample data corresponding to the labeled sample data based on a second auxiliary classification label of the original sample data corresponding to the labeled sample data. A method for determining a sample category set is configured to obtain a negative sample category set of original sample data corresponding to the labeled sample data, wherein the second auxiliary classification label is included in the original sample data. It is intended to indicate a category of targets that are not targeted.

In an optional embodiment, the second loss information determination module includes a classification label of the labeled sample data, frequency information corresponding to each category, and a positive sample category set of original sample data corresponding to the labeled sample data. And when determining the loss weight corresponding to each category of a plurality of predetermined categories for the labeled sample data based on the negative sample category set, for any one category, the labeled the classification label of the sample data indicates that the labeled sample data is non-background labeled sample data, the one category does not belong to the positive sample category set and the negative sample category set, and the configured to determine a loss weight corresponding to the one category as a first predetermined loss weight value if it is determined that frequency information corresponding to the one category is less than a predetermined frequency threshold; .

In an optional embodiment, the second loss information determination module includes a classification label of the labeled sample data, frequency information corresponding to each category, and a positive sample category set of original sample data corresponding to the labeled sample data. And when determining the loss weight corresponding to each category of a plurality of predetermined categories for the labeled sample data based on the negative sample category set, for any one category, the labeled the classification label of sample data indicates that the labeled sample data is background labeled sample data, and/or any one category belongs to the positive sample category set or the negative sample category set; and/or determining the loss weight corresponding to the one category as a second predetermined loss weight value when it is determined that the frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold. configured as

According to an eighth aspect, embodiments of the present application provide a data processing apparatus. The apparatus comprises a second processing module configured to perform data processing tasks based on a pre-trained data classification model, said data classification model according to any one of the third aspects above. wherein the data processing task includes one or more of data classification, data segmentation, instance segmentation, and the data is image data and text data including any one of

According to a ninth aspect, embodiments of the present application further provide a computer apparatus. The computer device comprises a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the computer device is operated, the processor and the memory communicates over a bus, and the processor executes the machine-readable instructions to perform the steps in any one possible embodiment of the first aspect or the first aspect above, or or performing the steps of any one possible embodiment of the second aspect, or performing the steps of the third or any one possible embodiment of the third aspect above, or the fourth aspect above or performing the steps in any one possible embodiment of the fourth aspect.

According to a tenth aspect, embodiments of the present application further provide a computer-readable storage medium. A computer program is stored on the computer-readable storage medium and, when the computer program is executed by a processor, instructs the processor to perform the steps in the first aspect or any one possible embodiment of the first aspect. or perform the steps of the second aspect or any one possible embodiment of the second aspect above; or perform the steps of the third aspect or any one possible embodiment of the third aspect above or cause the steps in the fourth aspect or any one possible embodiment of the fourth aspect to be performed.

According to an eleventh aspect, embodiments of the present application further provide a computer program product. The computer program, when executed by a processor, causes the processor to perform the steps of the first aspect or any one possible embodiment of the first aspect above, or the steps of any one of the second aspect or the second aspect above. or any one of the possible embodiments in the third or third aspect above, or any one of the fourth or fourth aspects above. or causes the steps in one possible embodiment to be performed.

In order to make the above objects, features and advantages of the present application clear and comprehensible, preferred embodiments are described in detail below with reference to the accompanying drawings.

Figure 4 is a flow chart illustrating a method for training an image classification model according to embodiments of the present application; 4 is a flow chart illustrating a specific method for obtaining loss information according to an embodiment of the present application; 4 is a flowchart illustrating a method for training a data classification model according to embodiments of the present application; 1 is a schematic diagram of an image classification model training device according to an embodiment of the present application; FIG. 1 is a schematic diagram of a data classification model training device according to an embodiment of the present application; FIG. 1 is a schematic diagram of a computing device according to an embodiment of the present application; FIG. FIG. 3 is a schematic diagram of another computing device according to embodiments of the present application;

In order to describe the technical solutions in the embodiments of the present application more clearly, the drawings required in the embodiments are briefly described below. The drawings attached hereto are taken into the specification and constitute a part of the specification, show the embodiments compatible with the application, and are used to interpret the technical solution of the application together with the specification. It should be understood that the following drawings depict several embodiments of the present application for purposes of illustration only and are not limiting of the present application. Those skilled in the art can also derive other related drawings based on these drawings without creative effort.

In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the following clearly and completely describes the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. . Of course, the described embodiments are only some embodiments of the present application rather than all embodiments. Generally, the units of the embodiments of the present application illustrated and shown in the drawings can be arranged and designed in a wide variety of different configurations. Accordingly, the following detailed description of embodiments of the present application provided with reference to the drawings is not intended to limit the scope of the claimed application, but is merely illustrative of embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by persons skilled in the art without creative efforts fall within the scope of protection of the present application.

According to the study, in the process of training a multi-category neural network model, for any one category, the positive samples of the other categories constitute the negative samples of the any one category, and the positive samples of the any one category If the number of positive samples is small, for any one category, the balance between positive and negative samples is lost, negative samples have a significant impact on any one category, and the multi-category neural network model is It has been found that this leads to low recognition accuracy when recognizing target objects in either category.

According to the above discussion, the present application provides a method for training an image classification model. When training the classification model, based on the labeled sample images and the classification labels respectively corresponding to each labeled sample image, frequency information respectively corresponding to each of the predetermined plurality of categories is determined. Then, based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category, loss information of the classification model is determined, and the classification model is trained based on the loss information. In the process, image classification is performed in the training process of an image classification model based on the frequency information of each category to determine the loss information according to the frequency information corresponding to each category and the classification label respectively corresponding to each sample image. You can adjust the degree of influence of positive and negative samples for a particular category on the model, which balances the influence of positive and negative samples on each category, and further refines the classification model for categories with few positive samples. can improve the recognition accuracy of

Any deficiencies existing in the above technical solutions are the results obtained by the inventors after practice and diligent study. Therefore, both the discovery process of the above problem and the solution proposed by the present application to the above problem in the following description should be the contribution that the inventors make to the present application in their disclosure.

The following clearly and completely describes the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Of course, the described embodiments are only some embodiments of the present application rather than all embodiments. Generally, the units of the embodiments of the present application illustrated and shown in the drawings can be arranged and designed in a wide variety of different configurations. Accordingly, the following detailed description of embodiments of the present application provided with reference to the drawings is not intended to limit the scope of the claimed application, but is merely illustrative of embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by persons skilled in the art without creative efforts fall within the scope of protection of the present application.

In the following drawings, similar symbols and letters represent similar elements, so that once an element is defined in one drawing, it need not be defined and explained in subsequent drawings. Please note.

In order to facilitate understanding of this embodiment, first, the training method of the image classification model disclosed in the embodiment of the present application will be described in detail. The execution body of the classification model training method provided in the embodiments of the present application is generally a computer device with a certain computing power. Such computer equipment includes, for example, terminal equipment, servers or other processing equipment. Terminal Equipment includes User Equipment (UE), Mobile Equipment, User Terminal, Terminal, Cellular Telephone, Cordless Telephone, Personal Digital Assistant (PDA), Handheld Equipment, Computing Equipment, Vehicle Equipment, Wearable Equipment. and so on. In some possible implementations, the classification model training method may be implemented by a processor invoking computer readable instructions stored in memory.

Example 1
As shown in FIG. 1, which is a flowchart illustrating training an image classification model according to an embodiment of the present application, the method includes steps S101-S104, wherein:
At S101, a labeled sample image set is obtained, and the labeled sample image set includes a plurality of labeled sample images and a classification label corresponding to each labeled sample image respectively.

In S102, based on the labeled sample images and the classification labels respectively corresponding to the respective labeled sample images, frequency information corresponding to each of the plurality of predetermined categories is determined.

In S103, the loss information of the image classification model is determined based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information corresponding to each category.

At S104, based on the loss information, an image classification model is trained, the image classification model is for determining categories of images to be recognized.

Each of S101 to S104 will be described in detail below.

I: In S101 above, a plurality of labeled sample images are included in the labeled sample image set. Different labeled sample images can be generated by labeling different original sample images, or can be generated by labeling the same original sample image. Each labeled sample image includes at least one labeled target object and tags corresponding to the target object. Here, the target object is any one of multiple objects that can be recognized by the image classification model, and the tag corresponding to the target object is the category to which the target object belongs.

For example, one original sample image includes not only the target object A but also the target object B. Therefore, when labeling the original sample image, the position of the target object A in the original sample image can be labeled and a classification label "target object A" added to form a labeled sample image. , the position of the target object B in the original sample image can also be labeled and a classification label “target object B” added to form another labeled sample image.

Note that if one original sample image contains multiple identical target objects, one labeled sample image can be formed for each target object. For example, if one original sample image contains three target objects A, when labeling the original sample image, for the first target object A, the first target object in the original sample image A, add the classification label "target object A"; for the second target object A, label the position of the second target object A in the original sample image; Add "target object A", for the third target object A, label the position of the third target object A in the original sample image, and add the classification label "target object A".

It should be noted that when multiple identical target objects are included in one original sample image, multiple labeled sample images can also be formed for all of the same target objects.

For example, when three target targets A are included in one original sample image, when labeling the original sample image, for three target targets A, three target targets A in the original sample image Label the location and add the tag "Target Subject A".

A specific labeling scheme may depend on the purpose of classification. For example, if the purpose of classification is instance segmentation, different labeled sample images are formed for different target objects. For example, if the purpose of classification is to recognize which images have a particular object, one labeled sample image can be formed for all the same target objects.

Before training an image classification model, first determine the categories of objects that the image classification model can recognize, and then, based on the determined categories of objects that the image classification model can recognize, create a set of labeled sample images. decide.

The labeled sample image set includes labeled sample images corresponding to each category, that is, for any one category, positive samples corresponding to the category are included.

For any one category, any labeled sample image corresponding to another category constitutes a negative sample of that one category.

Note that the labeled sample image set may further include a plurality of labeled sample images that are all negative samples for all categories.

For example, when labeling one original image, the positions of objects in the original sample image that do not belong to any one category are labeled, and the classification label "no category" or "background" is added.

In another example, the classification labels may be represented in numerical form. For example, if there are 30 target categories that can be recognized by the image classification model, 1 to 30 represent the 30 categories, and 0 is the classification label for labeled sample images that are negative samples for all categories. used as

II: In S102 above, the frequency information corresponding to any one category is, for example,
The number of labeled sample images corresponding to any one category, the ratio of labeled sample images corresponding to any one category to a set of labeled sample images, and the labeled sample images corresponding to any one category any one of the number of original sample images of sample images, and the ratio of the number of original sample images of labeled sample images corresponding to any one category to the number of original sample images of all labeled sample images can be one.

Furthermore, any one of the following methods (1), (2), and (3) can be used to determine frequency information corresponding to each category of a plurality of categories, but is not limited to these. .

(1) for each category, based on the labeled sample images belonging to the category and the original sample images corresponding to each labeled sample image, determine the number of original sample images corresponding to the category; The images correspond to one or more labeled sample images.

Frequency information corresponding to the category is determined based on the number of original sample images.

Here, the number of original sample images can be directly determined from the frequency information corresponding to the category, and the ratio of the number of original sample images to the total number of original sample images is the frequency information corresponding to the category. can also be determined.

Specifically, different labeled sample images may come from the same original sample image, so after inputting different labeled sample images from the same labeled sample image into an image classification model, With the image classification model parameters remaining constant, the feature data extracted for different labeled sample images are similar and thus identical. When training an image classification model using these labeled sample images, the features that an image classification model can learn are: Fewer features than the model can learn.

For example, the labeled sample image a1 and the labeled sample image a2 are from the same original sample image A, and the labeled sample image b1 is from the original sample image B. The image classification model can learn fewer features when performing feature learning on a1 and a2 than it can learn when performing feature learning on a1 and b1.

Furthermore, when determining the frequency information corresponding to each category based on the number of labeled sample images directly, it is assumed that more than one of the positive samples corresponding to one category are from the same original sample image. , even if the number of positive samples is large, the number of features corresponding to the category that the image classification model can learn is small, and the recognition accuracy for the category is low. It invites judgment with many categories. To avoid such misjudgment, in this embodiment, the frequency information is determined based on the number of original sample images corresponding to each labeled sample image.

(2) For each category, determine frequency information corresponding to the category based on the total number of labeled sample images in the labeled sample image set and the number of labeled sample images belonging to the category. The ratio of the number of labeled sample images belonging to the category to the total number of labeled sample images can be used as frequency information corresponding to the category.

(3) The number of labeled sample images belonging to the category is used as frequency information corresponding to the category.

Here, for one category, it is recognized that the higher the frequency information of the category, the greater the number of positive samples corresponding to the category, and the higher the balance between the positive samples and the negative samples of the category. It is recognized that the lower the frequency information, the lower the number of positive samples corresponding to that category, and the lower the balance between positive and negative samples in that category.

III: In the above S103 and S104, if the loss information of the image classification model is determined based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information corresponding to each category, for example,
For each labeled sample image, the loss when classifying the labeled sample image using the image classification model based on the classification label of the labeled sample image and the frequency information corresponding to each category respectively A method of obtaining information can be used.

In the specific implementation process, in the process of training an image classification model based on labeled sample images, a classification result can be determined each time a labeled sample image is input to the image classification model. The classification result is, for example, one vector. The number of elements in the vector matches and corresponds one-to-one with the number of categories of interest that the image classification model can recognize. The value of any one element in the vector represents the probability or confidence that the labeled sample image belongs to the category corresponding to that element.

For example, if the number of target categories that the image classification model can recognize is 30, after inputting any one labeled sample image into the image classification model, the resulting classification result is a 30-dimensional vector. , (p1, p2, ..., p30). Here, the value of the i-th element pi in the vector represents the degree of confidence that the labeled sample image belongs to the category corresponding to pi. more likely to belong to the category of

Loss information corresponding to the labeled sample images can then be obtained based on the vectors, classification labels corresponding to the labeled sample images, and frequency information respectively corresponding to each category.

Based on the loss information, the parameters of the image classification model are adjusted to tend to reduce the loss information, completing a one-time parameter adjustment process for the image classification model.

The next labeled sample image is then input to the image classification model and the loss information determination and parameter adjustment process described above is rerun.

After performing a single parameter tuning process for an image classification model using all labeled sample images, a single training for the image classification model is considered complete.

After training the image classification model multiple times, or when the loss information of the image classification model converges, a final trained image classification model can be obtained.

By way of example, as shown in FIG. 2, an embodiment of the present application uses a classification model to classify the labeled sample images based on the classification labels of the labeled sample images and the frequency information respectively corresponding to each category. We also provide a specific method for obtaining loss information when performing classification with The method includes:

In S201, based on the classification label of the labeled sample image and the frequency information corresponding to each category, a loss weight corresponding to each of a plurality of predetermined categories is determined for the labeled sample image. do.

In S202, based on the loss weight corresponding to each category of a plurality of predetermined categories, loss information is obtained when the labeled sample image is classified using the classification model.

（１） Here, for each labeled sample image, its loss information satisfies, for example, the following formula (1).

(1)

は、下記式（２）を満たす。

（２） here,

satisfies the following formula (2).

(2)

は、該ラベル付きサンプル画像がカテゴリｊに属する信頼度を表し、ｃは、該ラベル付きサンプル画像に対応する注釈ラベルを表し、

は、該ラベル付きサンプル画像に対して、カテゴリｊに対応する損失重みを表す。 where C represents the number of categories of interest that the classification model can recognize,

represents the confidence that the labeled sample image belongs to category j, c represents the annotation label corresponding to the labeled sample image,

represents the loss weight corresponding to category j for the labeled sample image.

In one embodiment, when determining loss weights, for example, for any one category, based on the classification label of the labeled sample image, the classification label of the labeled sample image is the is a background labeled sample image, and determining whether the frequency information corresponding to any one category is less than a predetermined frequency threshold.

the classified label of the labeled sample image indicates that the labeled sample image is a non-background labeled sample image, and the frequency information corresponding to any one category is less than a predetermined frequency threshold. If so, a loss weight corresponding to the one category is determined as the first predetermined loss weight value.

the classified label of the labeled sample image indicates that the labeled sample image is a background labeled sample image; and/or the frequency information corresponding to any one category is greater than or equal to a predetermined frequency threshold. If so, the loss weight corresponding to the one category is determined as the second predetermined loss weight value.

Here, the specific numerical value of the predetermined frequency threshold may be specifically set according to the form of frequency information (for example, number or ratio) and the requirements for classification accuracy.

Here, specific numerical values of the first predetermined loss weight value and the second predetermined loss weight value may be specifically set according to actual demand.

In the above embodiment, during the training process of the image classification model, the degree of influence of the positive and negative samples of a particular category on the image classification model can be adjusted, so that the detection accuracy of the classification model for categories with few positive samples improve. Accordingly, the first predetermined loss weight value is generally less than the second predetermined loss weight value.

In the process of training an image classification model using a labeled sample image a, if positive samples and negative samples of category m are unbalanced for category m, the frequency information corresponding to category m is While below a predetermined frequency threshold, the annotation label of the labeled sample image a indicates that the labeled sample image a belongs to one category, but the category is not category m, but also the background category. do not have. In this case, if category m is a category whose frequency information is less than the frequency threshold, reduce and even ignore the influence of the labeled sample image a on the category m, i.e. Reduce the influence of the labeled sample image a.

In the process of training an image classification model using the labeled sample image a, for category m, if the positive and negative samples of category m are not unbalanced, the labeled sample on category m The influence of image a cannot be reduced, nor can it be ignored.

In the process of training an image classification model using the labeled sample images a, for category m, the positive and negative samples of category m are not unbalanced, but the labeled sample images a are all If for a category all are labeled sample images of the background category, the influence of the labeled sample image a on the category m cannot be reduced or even ignored.

は、下記式（３）を満たす。

（３） For example, the first predetermined loss weight value is 0 and the second predetermined loss weight value is 1 if the influence of the labeled sample images on the category is ignored. The classification labels of the target category that can be recognized by the classification model can be set to 1 to C, and the classification label of the background category can be set to 0. Therefore, for one labeled sample image, the loss weight corresponding to category j

satisfies the following formula (3).

(3)

は、所定の頻度閾値を表し、ｃは、ラベル付きサンプル画像が属するカテゴリを表し、

は、ラベル付きサンプル画像が非背景のラベル付きサンプル画像であることを表す。

は、カテゴリｊに対応する頻度情報を表す。 here,

represents a predetermined frequency threshold, c represents the category to which the labeled sample image belongs,

indicates that the labeled sample image is a non-background labeled sample image.

represents the frequency information corresponding to category j.

は、下記式（４）を満たす。

（４） Also for example, the first predetermined loss weight value is 0.3 and the second predetermined loss weight value is 1 to reduce the influence of the labeled sample images on the category. Loss weight corresponding to category j for one labeled sample image

satisfies the following formula (4).

(4)

Here, when reducing the influence of labeled sample images on categories, the specific numerical values of the first predetermined loss weight value and the second predetermined loss weight value may be set according to actual needs. , for example, 0.1, 0.2, 0.4, 0.5, and so on. The second predetermined loss weight value may also be, for example, 0.9, 1.1, 1.2, and so on. not limited here.

It should be noted that the embodiments of the present application further provide another specific scheme for determining loss weights. In the embodiment, for example, based on the classification labels of the labeled sample images, frequency information corresponding to each category, and a positive sample category set and a negative sample category set of the original sample images corresponding to the labeled sample images, Loss weights corresponding to each category of a predetermined plurality of categories can be determined for the labeled sample images.

Here, the positive sample category set includes at least one target subject category included in the original sample image.

Specifically, either one of schemes 1 or 2 below can be used to determine the positive sample category set of the original sample images corresponding to the labeled sample images.

Method 1: from a set of labeled sample images, determine a target labeled sample image corresponding to the same original sample image as the labeled sample image;
Based on the classification labels respectively corresponding to the labeled sample images and the target labeled sample images, a positive sample category set of the original sample images corresponding to the labeled sample images is determined.

Here, when forming labeled sample images, a plurality of different labeled sample images may be labeled based on the same original sample image.

For example, for any one labeled sample image a, the corresponding target labeled sample images of the same original sample image are labeled sample image b, labeled sample image c, and labeled sample image d. be. Here, the classification label of the labeled sample image a is m2, the classification label of the labeled sample image b is m5, the classification label of the labeled sample image c is m9, and the labeled sample image d is m5. Therefore, the positive sample category set of the labeled sample image a is {m2, m5, m9}.

Method 2: Determine the positive sample category set of the original sample image corresponding to the labeled sample image according to the first auxiliary classification label of the original sample image corresponding to the labeled sample image.

Here, in the process of labeling the original sample images and generating labeled sample images, the number of labeled sample images to be generated is too large. are often unable to generate different labeled sample images. In this case, a first auxiliary classification label can be added to the original sample image. The first sub-classification label is for indicating the category of the target object contained in the original sample image.

Here, the first sub-classification label, when labeled, can indicate all categories of the target object included in the original sample image, and can indicate only some categories of the target object included in the original sample image. Note that you can also

For example, all categories of target objects contained in the original sample image include m1, m3, m7, m8, and m11, and if the original sample image is labeled with the first sub-classification label, the first sub-classification label are, for example, m1, m3, m7. In this case, the positive sample category set of the formed original sample image is {m1, m3, m7}.

The first sub-classification labels are, for example, m7, m8, and m11. In this case, the positive sample category set of the formed original sample image is {m7, m8, m11}.

The first auxiliary classification labels are, for example, m1, m3, m7, m8, and m11. In this case, the positive sample category set of the formed original sample image is {m1, m3, m7, m8, m11}.

A negative sample category set includes at least one target object category that is not included in the original sample image.

Now, Method 3 below can be used to determine the negative sample category set of the original sample image corresponding to the labeled sample image.

Method 3: Determine the negative sample category set of the original sample image corresponding to the labeled sample image according to the second auxiliary classification label of the original sample image corresponding to the labeled sample image.

Here, as in method 2 above, when labeling the original sample images, it is also possible to add a second auxiliary classification label to the original sample images. The second sub-classification label is for indicating a target object category that is not included in the original sample image.

For example, when labeling the original sample image, when it is determined that the target object category not included in the original sample image includes m3, m9, and m15, label the original sample image with a second auxiliary classification label. , the second sub-classification labels are, for example, m3, m9, and m15. In this case, the negative sample category set of the formed original sample image is {m3, m9, m15}.

Note that the content indicated by the first sub-classification label or the second sub-classification label may be null. In this case, the positive sample category set and the negative sample category set are also null.

In the embodiment, when determining a loss weight corresponding to each category of a plurality of predetermined categories for the labeled sample image, for example, for any one category, the labeled sample image that the classification label of the labeled sample image indicates whether the labeled sample image is a background labeled sample image, and the frequency corresponding to any one category Determining whether the information is below a predetermined frequency threshold and determining whether the any one category belongs to a positive sample category set and a negative sample category set corresponding to the labeled sample image.

the classification label of the labeled sample image indicates that the labeled sample image is a non-background labeled sample image, and the one category does not belong to a positive sample category set and a negative sample category set; and if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold, then determining the loss weight corresponding to the one category as a first predetermined loss weight value.

the classification label of the labeled sample image indicates that the labeled sample image is a background labeled sample image, and/or the any one category belongs to a positive sample category set or a negative sample category set. and/or determining the loss weight corresponding to the one category as a second predetermined loss weight value when it is determined that the frequency information corresponding to the one category is greater than or equal to a predetermined frequency threshold. do.

In the embodiment, when adjusting the degree of influence of positive and negative samples of any category on the image classification model, in the process of reducing or ignoring the influence of labeled sample image a on category m, If the target object of the category m always exists in the original sample image corresponding to the labeled sample image a, the influence of the labeled sample image on the category m is not ignored. Therefore, the image classification model can learn more features that fall under the category, and the accuracy of the image classification model can be further improved.

Note that if the original sample image does not necessarily have a target object of a particular category, then the original sample image is in fact recognized as the background image of the category m, and based on the original sample image on the category m Do not ignore the influence of the resulting labeled sample images. Therefore, the image classification model can be trained on the differential features of the target object of category m, and the accuracy of the image classification model can be further improved.

Further, in an embodiment of the present application, it is further determined whether the any one category belongs to a positive sample category set or a negative sample category set corresponding to the labeled sample images. If so, the influence of the labeled sample image a on the category m cannot be reduced or even ignored.

は、下記式（５）を満たす。

（５） For example, the first predetermined loss weight value is 0 and the second predetermined loss weight value is 1 if the influence of the labeled sample images on the category is ignored. Therefore, for one labeled sample image, the loss weight corresponding to category j

satisfies the following formula (5).

(5)

は、所定の頻度閾値を表し、

は、ラベル付きサンプル画像が非背景のラベル付きサンプル画像であることを表す。

は、カテゴリｊに対応する頻度情報を表す。

は、ポジティブサンプルカテゴリ集合を表し、

は、ネガティブサンプルカテゴリ集合を表す。 here,

represents a predetermined frequency threshold,

indicates that the labeled sample image is a non-background labeled sample image.

represents the frequency information corresponding to category j.

represents the positive sample category set, and

represents the negative sample category set.

は、下記式（６）を満たす。

（６） Also for example, to reduce the influence of the labeled sample images on the category, the first predetermined loss weight value is 0.15 and the second predetermined loss weight value is 1.1. Loss weight corresponding to category j for one labeled sample image

satisfies the following formula (6).

(6)

When reducing the influence of the labeled sample images on the category, the specific numerical values of the first predetermined loss weight value and the second predetermined loss weight value may be set according to actual needs.

Embodiments of the present application, when training a classification model, based on the labeled sample images and the classification labels corresponding to each labeled sample image, respectively, generate frequency information respectively corresponding to each of a plurality of predetermined categories. decide. Then, based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category, loss information of the classification model is determined, and the classification model is trained based on the loss information. In the process, image classification is performed in the training process of an image classification model based on the frequency information of each category to determine the loss information according to the frequency information corresponding to each category and the classification label respectively corresponding to each sample image. You can adjust the degree of influence of positive and negative samples for a particular category on the model, which balances the influence of positive and negative samples on each category, and further refines the classification model for categories with few positive samples. can improve the recognition accuracy of

Embodiments of the present application further provide another image processing method. The method includes performing an image processing task based on a pre-trained image classification model;
The image classification model is trained by the image classification model training method of the above embodiment,
Image processing tasks include one or more of image classification, object detection, keypoint detection, image segmentation, instance segmentation.

In the embodiments of the present application, the image classification model can be specifically selected based on the actual image processing task. For example, if the image processing task is object detection, a faster Region-Convolutional Neural Network (Faster R-CNN) can be used as the image classification model. Also, for example, when the image processing task is instance segmentation, the instance segmentation model Mask R-CNN can be used as the image classification model.

In this example, the labeled sample sets for training the image classification model are Microsoft Common Objects in Context (MS COCO), A Dataset for Large Vocabulary Instance Segmentation pattern analysis, statistical modeling and computational learning visual object classes (Pascal VOC), and Open Image datasets. or a sample set obtained after processing (eg, relabeling) at least one of the sample sets.

Embodiments of the present application, when performing image processing tasks, can be implemented by the image classification model trained by the image classification model training method disclosed in the above embodiments, and can obtain image processing results with higher accuracy. .

As shown in FIG. 3, embodiments of the present application further provide a method for training a data classification model. The method includes:

At S301, a labeled sample data set is obtained, and the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image.

In S302, based on the labeled sample data and the classification labels respectively corresponding to each labeled sample data, frequency information corresponding to each of the plurality of predetermined categories is determined.

In S303, the loss information of the data classification model is determined based on the classification labels respectively corresponding to the plurality of labeled sample data and the frequency information respectively corresponding to each category.

At S304, based on the loss information, a data classification model is trained, the data classification model is for determining categories of data to be recognized.

In such embodiments, the sample data set may be any one of a sample image set and a sample text set. Note that the labeled sample data differs depending on the data to be processed. Specifically, it depends on the actual application scene. Here, detailed description is omitted.

S101 to S104 in the above embodiment can be referred to for the specific implementation of S301 to S304, and the detailed description is omitted here.

Embodiments of the present application further provide methods of data processing. The method comprises
performing a data processing task based on a pre-trained data classification model;
The data classification model is trained by the data classification model training method in the above example,
the data processing task includes one or more of data classification, data segmentation, instance segmentation;
The data includes any one of image data and text data.

In the above method of specific embodiments, the description order of each step means a strict execution order and does not limit the implementation process in any way. Those skilled in the art should understand that logic dictates.

According to the same technical idea, the embodiments of the present application further provide an image classification model training device corresponding to the image classification model training method. Since the principle for solving the problem by the apparatus in the embodiments of the present application is similar to the training method of the above image classification model in the embodiments of the present application, the implementation of the apparatus can refer to the implementation of the method, and the duplicated Description is omitted.

As shown in FIG. 4, which is a schematic diagram of an image classification model training device according to an embodiment of the present application, said device comprises a first acquisition module 41, a first frequency information determination module 42 and a first loss information determination module 43. and a first training module 44;
the first obtaining module 41 is configured to obtain a set of labeled sample images, wherein the set of labeled sample images includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image;
A first frequency information determination module 42 determines frequency information respectively corresponding to each of a plurality of predetermined categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images. configured as
the first loss information determination module 43 is configured to determine loss information of an image classification model based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category;
A first training module 44 is configured to train the image classification model based on the loss information, the image classification model for determining categories of images to be recognized.

The description of the processing flow of each module and the interaction flow between each module in the apparatus can refer to the related descriptions in the above method embodiments, and detailed descriptions are omitted herein.

Another embodiment of the present application further provides an image processing device. The device is
a first processing module configured to perform an image processing task based on a pre-trained image classification model;
The image classification model is trained by the image classification model training method described in the above example,
The image processing tasks include one or more of image classification, object detection, keypoint detection, image segmentation, instance segmentation.

As shown in FIG. 5, which is a schematic diagram of a data classification model training device according to an embodiment of the present application, said device comprises a second acquisition module 51, a second frequency information determination module 52, and a second loss information determination module 53. and a second training module 54;
the second obtaining module 51 is configured to obtain a labeled sample data set, wherein the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image;
A second frequency information determination module 52 determines frequency information respectively corresponding to each of a plurality of predetermined categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data. configured as
a second loss information determination module 53 is configured to determine loss information of a data classification model based on the classification labels respectively corresponding to the plurality of labeled sample data and the frequency information respectively corresponding to each category;
A second training module 54 is configured to train the data classification model based on the loss information, the data classification model for determining categories of data to be recognized.

The description of the processing flow of each module and the interaction flow between each module in the apparatus can refer to the related descriptions in the above method embodiments, and detailed descriptions are omitted herein.

Embodiments of the present application further provide a data processing apparatus. The device is
a second processing module configured to perform a data processing task based on the pre-trained data classification model;
The data classification model is trained by the data classification model training method described in the above example,
The data processing task includes one or more of data classification, data segmentation, instance segmentation, and the data includes any one of image data and text data.

Embodiments of the present application further provide computing equipment 60 . As shown in FIG. 6, which is a schematic diagram showing the structure of computer equipment 60 according to an embodiment of the present application, computer equipment 60 comprises processor 61 , memory 62 and bus 63 . Machine readable instructions executable by the processor 61 in the memory 62 (for example, the first acquisition module 41, the first frequency information determination module 42, the first loss information determination module 43 and the first training module in the apparatus shown in FIG. 4). 44) are stored, and when computer equipment 60 is operated, said processor 61 and said memory 62 communicate via bus 63 and said machine readable instructions are executed by said processor 61. when
obtaining a labeled sample image set, wherein the labeled sample image set includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image;
Determining frequency information corresponding to each of a plurality of predetermined categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images;
Determining loss information for an image classification model based on the classification labels respectively corresponding to the plurality of labeled sample images and the frequency information respectively corresponding to each category;
Based on the loss information, training the image classification model, the image classification model for determining categories of images to be recognized.

For details of how the processor 61 executes each process, refer to the above embodiments. Here, detailed description is omitted.

Embodiments of the present application further provide computing equipment 70 . As shown in FIG. 7, which is a schematic diagram illustrating the structure of a computing device 70 according to an embodiment of the present application, the computing device 70 comprises a processor 71 , a memory 72 and a bus 73 . Machine readable instructions executable by the processor 71 in the memory 72 (for example, the second acquisition module 51, the second frequency information determination module 52, the second loss information determination module 53 and the second training module in the apparatus shown in FIG. 5). 54), and when computer equipment 70 is operated, said processor 11 and said memory 72 communicate via bus 73 and said machine readable instructions are executed by said processor 71. when
obtaining a labeled sample image set, wherein the labeled sample image set includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image;
obtaining a labeled sample data set, wherein the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image;
Determining frequency information respectively corresponding to each of a plurality of predetermined categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data;
Determining loss information for a data classification model based on the classification labels respectively corresponding to the plurality of labeled sample data and the frequency information respectively corresponding to each category;
Based on the loss information, training the data classification model, the data classification model for determining categories of data to be recognized.

For details of how the processor 71 executes each process, refer to the above embodiments. Here, detailed description is omitted.

Embodiments of the present application further provide a computer-readable storage medium. A computer program is stored on the computer readable storage medium, and when the computer program is executed by a processor, the method of training an image classification model, the method of image processing, the method of training a data classification model according to the above method embodiments. , or perform the steps of the data processing method. Here, the storage medium may be a volatile or non-volatile computer-readable storage medium.

A computer program product of an image classification model training method, an image processing method, a data classification model training method, or a data processing method provided by the embodiments of the present application includes a computer-readable storage medium storing program code. The instructions contained in the program code are used to perform the steps of the image classification model training method, the image processing method, the data classification model training method, or the data processing method described in the above method embodiments. Specifically, reference can be made to the embodiments of the above method, and the detailed description is omitted here.

Embodiments of the present application further provide computer programs. The computer program implements the method of any one of the above embodiments when executed by a processor. The computer program product may be specifically implemented in hardware, software or a combination thereof. In one alternative embodiment, said computer program product is specifically embodied as a computer storage medium, and in another alternative embodiment, said computer program product is specifically embodied as, for example, It is embodied as a software product such as a software development kit (SDK).

For the convenience and simplification of explanation, the specific working steps of the above-described systems and devices can refer to the corresponding steps in the method embodiments, so they will not be described in detail here. should be clearly understood by those skilled in the art. It should be understood that the systems, devices and methods disclosed in some of the embodiments provided by the present invention can be implemented in other manners. For example, the embodiments of the apparatus described above are merely exemplary, for example, the division of the units is merely the division of logic functions, and other division schemes may be used when actually implemented. For example, multiple units or assemblies may be combined or incorporated into another system. Or some features may be ignored or not implemented. Also, the mutual couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some communication interface, device or unit, electrical, mechanical or Other forms are also possible.

The units described as separate members may or may not be physically separate. Members shown as units may or may not be physical units. That is, they may be co-located or distributed over multiple networks. Some or all of these units can achieve the purpose of the measures of the present embodiment according to actual needs.

Also, each functional unit in each embodiment of the present application may be integrated into one processing unit, each unit may exist as a physically separate entity, and two or more units may be integrated into one unit. may be

The functions may be implemented in the form of software functional units and stored in a processor-executable volatile or non-volatile computer readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present application is essentially or the part that contributed to the prior art or part of the technical solution is embodied in the form of a software product. Such computer software products may be stored on a storage medium, and may be stored on a single computer device (such as a personal computer, server, or network device) to perform the methods described in each embodiment of the present application. contains some instructions for executing all or part of the steps of The above-mentioned storage media include USB memory, removable hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk, optical disk, etc. Various types of memory that can store program code. Including media.

Finally, I would like to mention that the above examples are only specific embodiments of the present application, and are for the purpose of describing the technical solutions of the present application, not limiting them, and The scope of protection is not limited to this. Although the present application has been described in detail with reference to the above embodiments, any person skilled in the art who is familiar with the technical field can understand the technical solutions described in the above embodiments within the technical scope disclosed in the present application. Modifications can be made to or changes can be easily conceived, or some technical features can be replaced by equivalents, and these modifications, changes or replacements are It should be understood that the essence of the technical solution to be used is not departed from the spirit and scope of the technical solution in the embodiments of the present application, but shall fall within the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims (15)

A method for training an image classification model performed by a computer device, comprising:
obtaining a labeled sample image set, wherein the labeled sample image set includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image;
Determining frequency information corresponding to each of a plurality of predetermined categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images;
for each of the labeled sample images, based on the classification label of the labeled sample image and the frequency information corresponding to each category, each of a plurality of predetermined categories for the labeled sample image; determining loss weights corresponding to categories; and loss in classifying the labeled sample images using a classification model based on the loss weights corresponding to each category of a plurality of predetermined categories. getting information and
training the image classification model based on the loss information, the image classification model for determining categories of images to be recognized;
Determining a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample image based on the classification label of the labeled sample image and frequency information corresponding to each category. teeth,
Based on the classification label of the labeled sample image, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample image corresponding to the labeled sample image, for the labeled sample image determining a loss weight corresponding to each category of a predetermined plurality of categories,
wherein the positive sample category set includes at least one target target category included in the original sample image, and the negative sample category set includes at least one target target category not included in the original sample image includes
Based on the classification label of the labeled sample image, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample image corresponding to the labeled sample image, to the labeled sample image In contrast, determining a loss weight corresponding to each category of a given plurality of categories is
for any one category
The classification label of the labeled sample image indicates that the labeled sample image is a non-background labeled sample image, and any one category belongs to the positive sample category set and the negative sample category set. first, if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold, determining the loss weight corresponding to the one category as a first predetermined loss weight value; including
How to train an image classification model. Determining frequency information corresponding to each of a plurality of predetermined categories based on the labeled sample images and classification labels respectively corresponding to each of the labeled sample images,
for each category, based on the labeled sample images belonging to the category and the original sample images corresponding to each labeled sample image, determining the number of original sample images corresponding to the category; the sample images correspond to one or more labeled sample images;
determining frequency information corresponding to the category based on the number of original sample images;
or for each category, determining frequency information corresponding to the category based on the total number of labeled sample images in the set of labeled sample images and the number of labeled sample images belonging to the category; 2. The method of training an image classification model according to claim 1, wherein the number of labeled sample images belonging to a category is used as frequency information corresponding to the category. Determining a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample image based on the classification label of the labeled sample image and frequency information corresponding to each category. teeth,
For any one category, the classification label of the labeled sample image indicates that the labeled sample image is a non-background labeled sample image, and frequency information corresponding to the any one category. is less than a predetermined frequency threshold, determining a loss weight corresponding to the one category as a first predetermined loss weight value; and/or for the one category, The classified label of the labeled sample image indicates that the labeled sample image is a background labeled sample image, and/or the frequency information corresponding to any one category is greater than or equal to a predetermined frequency threshold. 3. The method of claim 1 or 2, further comprising: determining a loss weight corresponding to the one category as a second predetermined loss weight value when determining . A training method for the image classification model includes:
determining a target-labeled sample image corresponding to the same original sample image as the labeled sample image from the set of labeled sample images, and assigning classification labels corresponding to the labeled sample image and the target-labeled sample image, respectively; determining a positive sample category set of the original sample image corresponding to the labeled sample image, or
A method for determining a positive sample category set of an original sample image corresponding to the labeled sample image based on a first auxiliary classification label of the original sample image corresponding to the labeled sample image, wherein the first auxiliary classification Labels are for indicating categories of target objects contained in the original sample image, using a scheme:
obtaining a positive sample category set of original sample images corresponding to the labeled sample images;
and/or the method of training the image classification model comprising:
Using a method of determining a negative sample category set of an original sample image corresponding to the labeled sample image based on a second auxiliary classification label of the original sample image corresponding to the labeled sample image, the labeled sample image obtaining a negative sample category set of the original sample image corresponding to the second sub-classification label for indicating categories of target objects not included in the original sample image. The method for training an image classification model according to any one of claims 1-3. Based on the classification label of the labeled sample image, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample image corresponding to the labeled sample image, to the labeled sample image In contrast, determining a loss weight corresponding to each category of a given plurality of categories is
for any one category, the classification label of the labeled sample image indicates that the labeled sample image is a background labeled sample image; and/or
any one category belongs to the positive sample category set or the negative sample category set, and/or
If it is determined that the frequency information corresponding to any one category is equal to or greater than a predetermined frequency threshold,
The method of claim 1 , further comprising determining a loss weight corresponding to the one category as a second predetermined loss weight value. An image processing method performed by a computer device, comprising:
performing an image processing task based on the pre-trained image classification model;
The image classification model is trained by the image classification model training method according to any one of claims 1 to 5 ,
The image processing method, wherein the image processing task includes one or more of image classification, object detection, keypoint detection, image segmentation, instance segmentation. A method for training a data classification model executed by a computer device, comprising:
obtaining a labeled sample data set, wherein the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image;
Determining frequency information respectively corresponding to each of a plurality of predetermined categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data;
For each labeled sample data, based on the classification label of the labeled sample data and the frequency information respectively corresponding to each category, each category of a plurality of predetermined categories for the labeled sample data determining loss weights corresponding to
Obtaining loss information when classifying the labeled sample data using the data classification model based on a loss weight corresponding to each category of a plurality of predetermined categories;
training the data classification model based on the loss information, the data classification model for determining categories of data to be recognized;
including
Determining a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample data based on the classification label of the labeled sample data and frequency information corresponding to each category. teeth,
Based on the classification label of the labeled sample data, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample data corresponding to the labeled sample data, for the labeled sample data determining a loss weight corresponding to each category of a predetermined plurality of categories,
wherein the positive sample category set includes at least one target category included in the original sample data, and the negative sample category set includes at least one target category not included in the original sample data includes
Based on the classification label of the labeled sample data, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample data corresponding to the labeled sample data, to the labeled sample data In contrast, determining a loss weight corresponding to each category of a given plurality of categories is
for any one category
The classification label of the labeled sample data indicates that the labeled sample data is non-background labeled sample data, and any one category belongs to the positive sample category set and the negative sample category set. first, if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold, determining the loss weight corresponding to the one category as a first predetermined loss weight value; How to train a data classification model, including A data processing method performed by a computer device, comprising:
performing a data processing task based on a pre-trained data classification model;
The data classification model is trained by the data classification model training method according to claim 7 ,
the data processing task includes one or more of data classification, data segmentation, instance segmentation;
A data processing method, wherein the data includes any one of image data and text data. An image classification model training device comprising:
a first acquisition module configured to acquire a set of labeled sample images, wherein the set of labeled sample images includes a plurality of labeled sample images and a classification label respectively corresponding to each labeled sample image; a first acquisition module;
First frequency information configured to determine frequency information respectively corresponding to each category of a predetermined plurality of categories based on the labeled sample images and classification labels respectively corresponding to each labeled sample image. a decision module;
For each labeled sample image, based on the classification label of the labeled sample image and the frequency information respectively corresponding to each category, each category of a plurality of predetermined categories for the labeled sample image and loss information when classifying the labeled sample image using an image classification model based on the loss weight corresponding to each category of a plurality of predetermined categories. a first loss information determination module configured to obtain;
a first training module configured to train the image classification model based on the loss information, the image classification model for determining categories of images to be recognized; 1 training module;
The first loss information determination module assigns the labeled sample image to each of a plurality of predetermined categories based on the classification label of the labeled sample image and frequency information corresponding to each category. When determining the corresponding loss weights,
Based on the classification label of the labeled sample image, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample image corresponding to the labeled sample image, for the labeled sample image determining a loss weight corresponding to each category of a predetermined plurality of categories, wherein said positive sample category set includes at least one target subject category included in said original sample image; wherein the negative sample category set includes at least one target subject category not included in the original sample image ;
The first loss information determination module is based on the classification labels of the labeled sample images, frequency information respectively corresponding to each category, a positive sample category set and a negative sample category set of the original sample images corresponding to the labeled sample images. to determine a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample image,
for any one category
The classification label of the labeled sample image indicates that the labeled sample image is a non-background labeled sample image, and any one category belongs to the positive sample category set and the negative sample category set. first, if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold, then determining the loss weight corresponding to the one category as a first predetermined loss weight value; An image classification model training device, comprising : An image processing device,
a first processing module configured to perform an image processing task based on a pre-trained image classification model;
The image classification model is trained by the image classification model training method according to any one of claims 1 to 5 ,
The image processing apparatus, wherein the image processing tasks include one or more of image classification, object detection, keypoint detection, image segmentation, instance segmentation. A data classification model training device comprising:
a second acquisition module configured to acquire a labeled sample data set, wherein the labeled sample data set includes a plurality of labeled sample data and a classification label respectively corresponding to each labeled sample image; a second acquisition module;
Second frequency information configured to determine frequency information respectively corresponding to each of a plurality of predetermined categories based on the labeled sample data and classification labels respectively corresponding to each of the labeled sample data. a decision module;
For each labeled sample data, based on the classification label of the labeled sample data and the frequency information respectively corresponding to each category, each category of a plurality of predetermined categories for the labeled sample data and determining loss information when classifying the labeled sample data using a data classification model based on the loss weight corresponding to each category of a plurality of predetermined categories. a second loss information determination module configured to obtain
a second training module configured to train the data classification model based on the loss information, the data classification model for determining categories of data to be recognized; 2 training modules;
The second loss information determination module, based on the classification label of the labeled sample data and frequency information respectively corresponding to each category, assigns the labeled sample data to each of a plurality of predetermined categories. When determining the corresponding loss weights,
Based on the classification label of the labeled sample data, the frequency information corresponding to each category, and the positive sample category set and negative sample category set of the original sample data corresponding to the labeled sample data, for the labeled sample data is configured to determine a loss weight corresponding to each category of a predetermined plurality of categories,
wherein the positive sample category set includes at least one target category included in the original sample data, and the negative sample category set includes at least one target category not included in the original sample data includes
The second loss information determination module is based on a classification label of the labeled sample data, frequency information corresponding to each category, and a positive sample category set and a negative sample category set of original sample data corresponding to the labeled sample data. to determine a loss weight corresponding to each of a plurality of predetermined categories for the labeled sample data,
for any one category
The classification label of the labeled sample data indicates that the labeled sample data is non-background labeled sample data, and any one category belongs to the positive sample category set and the negative sample category set. first, if it is determined that the frequency information corresponding to the one category is less than a predetermined frequency threshold, then determining the loss weight corresponding to the one category as a first predetermined loss weight value; A data classification model trainer, comprising : A data processing device,
a second processing module configured to perform a data processing task based on the pre-trained data classification model;
The data classification model is trained by the data classification model training method according to claim 7 ,
the data processing task includes one or more of data classification, data segmentation, instance segmentation;
The data processing device, wherein the data includes any one of image data and text data. A computer device, said computer device comprising a processor, a memory, and a bus, wherein said memory stores machine-readable instructions executable by said processor, and when the computer device operates: The processor and the memory communicate via a bus, and the processor executes the machine-readable instructions to perform the steps of the method for training an image classification model according to any one of claims 1-5 . or perform the steps of the image processing method of claim 6 , or perform the steps of the data classification model training method of claim 7 , or perform the data processing of claim 8 A computer device that performs the steps of a method. A computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the processor executes the program according to any one of claims 1 to 5 Execute the steps of the image classification model training method according to claim 6, or execute the steps of the image processing method according to claim 6 , or execute the steps of the data classification model training method according to claim 7 . , or a computer readable storage medium for performing the steps of the data processing method according to claim 8 . A computer program which, when executed by a processor, causes said processor to perform the steps of the method for training an image classification model according to any one of claims 1 to 5 , or Execution of the steps of the image processing method according to claim 6 , or execution of the steps of the data classification model training method of claim 7 , or execution of the steps of the data processing method according to claim 8 . , a computer program.

Images