Commodity attribute detection method and device, storage medium and computer device
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- VIPSHOP (GUANGZHOU) SOFTWARE CO LTD
- Filing Date
- 2022-12-28
- Publication Date
- 2026-05-12
AI Technical Summary
Existing product attribute detection methods have low accuracy, mainly because they cannot effectively determine whether product images are suitable for attribute detection, resulting in a large deviation between the detection results and the actual situation.
A pose detection model and a binary classification model are introduced. The pose detection model outputs pose key point data, and the binary classification model is used to determine whether each product image is suitable for attribute detection, and suitable images are selected for attribute detection.
It improves the accuracy of product attribute detection, eliminates interference from inappropriate product images, and enhances the accuracy of detection results.
Smart Images
Figure CN115937603B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence, and in particular to a method, apparatus, storage medium, and computer equipment for detecting product attributes. Background Technology
[0002] In e-commerce platforms, to facilitate users' quick understanding of products and to provide personalized recommendations, it's necessary to determine the attribute values for each product's various attributes. Taking clothing as an example, clothing products may include attributes such as fit, length, style, and sleeve length, which users can use to quickly understand the product. However, existing product attribute detection methods suffer from low accuracy. Summary of the Invention
[0003] The purpose of this application is to address at least one of the aforementioned technical deficiencies, particularly the low detection accuracy in the prior art.
[0004] In a first aspect, embodiments of this application provide a product attribute detection method, the method comprising:
[0005] Obtain multiple product images of the product to be inspected;
[0006] Obtain the pose detection model and the binary classification model;
[0007] Each of the product images is input into the pose detection model to obtain the pose key point data corresponding to each product image;
[0008] Based on the preset image filtering rules and the pose key point data corresponding to each of the product images, the first target image is selected from each of the product images;
[0009] The pose key point data corresponding to each of the first target images are input into the binary classification model to obtain the classification result of each first target image. Each classification result is used to reflect whether the corresponding first target image is suitable for attribute detection.
[0010] Based on the classification results of each of the first target images, a second target image suitable for attribute detection is selected from each of the first target images;
[0011] Based on each of the second target images, the predicted product attribute values of the product to be detected are detected.
[0012] In one embodiment, the step of detecting the predicted product attribute value of the product to be detected based on each of the second target images includes:
[0013] Determine the attribute to be detected and obtain the single-label multi-classification model corresponding to the attribute to be detected;
[0014] Each of the second target images is input into the single-label multi-classification model to obtain the detection attribute values output by the single-label multi-classification model and the confidence level corresponding to each detection attribute value;
[0015] Based on each of the detected attribute values and the confidence level corresponding to each detected attribute value, the predicted product attribute value is determined among the detected attribute values.
[0016] In one embodiment, the product attribute detection method further includes:
[0017] Obtain the initial product attribute values of the product to be tested, which are pre-labeled by humans;
[0018] If the initial product attribute value is different from the predicted product attribute value, and the confidence level corresponding to the predicted product attribute value is greater than the preset confidence level threshold, then attribute modification information is pushed to the information maintenance personnel.
[0019] In one embodiment, the single-label multi-classification model is trained using the Focal Loss function as the loss function.
[0020] In one embodiment, the step of selecting the first target image from the product images based on preset image filtering rules and the pose key point data corresponding to each product image includes:
[0021] Based on the pose key point data corresponding to each product image, it is determined whether each product image is a model display image, and each model display image is used as the first target image.
[0022] In one embodiment, the step of obtaining the binary classification model includes:
[0023] Obtain an initial linear regression model and a training dataset, wherein the training dataset includes multiple training images and pre-labeled manual classification results for each training image;
[0024] Each training image is input into the pose detection model to obtain pose key point data corresponding to each training image;
[0025] The pose key point data corresponding to each training image are input into the initial linear regression model to obtain the training classification result corresponding to each training image. Based on the training classification result and the manual classification result, the initial linear regression model is iteratively trained until the preset training completion condition is met and the binary classification model is obtained.
[0026] Secondly, embodiments of this application provide a product attribute detection device, the device comprising:
[0027] The product image acquisition module is used to acquire multiple product images of the product to be inspected.
[0028] The binary classification model acquisition module is used to acquire the pose detection model and the binary classification model.
[0029] The pose detection module is used to input each of the product images into the pose detection model to obtain the pose key point data corresponding to each product image.
[0030] The first target image filtering module is used to filter out the first target image from each of the product images according to the preset image filtering rules and the posture key point data corresponding to each of the product images;
[0031] The classification module is used to input the pose key point data corresponding to each of the first target images into the binary classification model to obtain the classification result of each first target image. Each classification result is used to reflect whether the corresponding first target image is suitable for attribute detection.
[0032] The second target image filtering module is used to filter out second target images suitable for attribute detection from each of the first target images based on the classification results of each first target image.
[0033] The attribute value detection module is used to determine the predicted product attribute value of the product to be detected based on each of the second target images.
[0034] In one embodiment, the attribute value detection module includes:
[0035] A single-label multi-classification model acquisition unit is used to determine the attribute to be detected and acquire the single-label multi-classification model corresponding to the attribute to be detected.
[0036] The detection unit is used to input each of the second target images into the single-label multi-classification model to obtain each detection attribute value output by the single-label multi-classification model and the confidence level corresponding to each detection attribute value;
[0037] An attribute value determination unit is used to determine the predicted product attribute value among the detected attribute values based on each of the detected attribute values and the confidence level corresponding to each of the detected attribute values.
[0038] Thirdly, embodiments of this application provide a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the commodity attribute detection method described in any of the above embodiments.
[0039] Fourthly, embodiments of this application provide a computer device, including: one or more processors, and a memory;
[0040] The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the commodity attribute detection method described in any of the above embodiments.
[0041] The product attribute detection method, apparatus, storage medium, and computer equipment of this application introduce a pose detection model and a binary classification model to perform pose detection on each product image. Based on the pose keypoint data output by the pose detection model and the binary classification model, it is determined whether each product image is suitable for attribute detection. This allows for the selection of suitable product images from multiple images for attribute detection, and the subsequent attribute detection. This eliminates interference from unsuitable product images, thereby improving the accuracy of attribute detection. Attached Figure Description
[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0043] Figure 1 This is one of the flowcharts illustrating a product attribute detection method in one embodiment;
[0044] Figure 2 This is a flowchart illustrating the steps for obtaining a binary classification model in one embodiment;
[0045] Figure 3 This is a flowchart illustrating the steps of detecting the predicted product attribute values of the product to be detected based on each of the second target images in one embodiment.
[0046] Figure 4 This is a second flowchart illustrating a product attribute detection method in one embodiment;
[0047] Figure 5 This is a schematic diagram of the product attribute detection device in one embodiment;
[0048] Figure 6 This is a schematic diagram of the structure of a computer device in one embodiment. Detailed Implementation
[0049] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0050] As mentioned in the background section, existing product attribute detection methods suffer from low accuracy. The inventors discovered that this problem arises because, in product attribute detection based on product images, multiple images of the same product exist, each with different display methods and content. Therefore, some product images are unsuitable for attribute detection. For example, a single garment product may have model display images, partial display images, and fabric display images. Model display images may include multiple images in different poses, such as upright and seated poses. However, existing technologies cannot determine whether each product image is suitable for detection; they can only randomly select any product image or use all product images for attribute detection. This leads to significant deviations between the attribute detection results and the actual situation, resulting in low accuracy.
[0051] To address the aforementioned problems, this application provides a product attribute detection method, apparatus, storage medium, and computer equipment. By introducing a pose detection model and a binary classification model to process each product image separately, it is possible to determine whether each product image is suitable for attribute detection. This allows for the selection of suitable product images from among the available images for attribute detection. In this way, interference from unsuitable product images can be eliminated, thereby improving the accuracy of attribute detection.
[0052] In one embodiment, this application provides a product attribute detection method. The following embodiment uses the application of this method to a computer device as an example for illustration. It can be understood that the computer device can be a device with data processing capabilities, and can be, but is not limited to, a personal laptop, desktop computer, a single server, or a server cluster, etc. Figure 1 As shown, the method may include the following steps:
[0053] S102: Obtain multiple product images of the product to be inspected.
[0054] In this context, "products to be inspected" refers to goods whose attributes are to be detected, such as clothing. On e-commerce platforms, suppliers can upload multiple product images of the same item to showcase its details. In this step, the computer equipment can acquire multiple product images of the same item.
[0055] S104: Obtain the pose detection model and the binary classification model.
[0056] Among them, the pose detection model refers to the model used to detect human pose. This model can perform pose detection on the input image and output pose key point data, so as to represent the human pose in the input image through the pose key point data.
[0057] A binary classification model is a model used to determine whether an input image is suitable for attribute detection. Specifically, a binary classification model receives an input image and outputs a classification result for that input image. This classification result can be used to reflect whether the input image is suitable for attribute detection; for example, the classification result can be "suitable" or "unsuitable".
[0058] In one embodiment, such as Figure 2 As shown, the steps for obtaining the binary classification model include:
[0059] S202: Obtain the initial linear regression model and training dataset, wherein the training dataset includes multiple training images and pre-labeled manual classification results for each training image.
[0060] The initial linear regression model refers to an untrained linear regression model. Multiple training images can be model displays in different poses and from different angles, such as standing or sitting poses. Each training image corresponds to a manually labeled result, which indicates whether the training image is suitable for attribute detection.
[0061] S204: Input each of the training images into the pose detection model to obtain the pose key point data corresponding to each training image.
[0062] In this step, for each training image, the computer device can input the training image into the pose detection model to perform pose detection on the training image and output the pose key point data corresponding to the training image to represent the human pose in the training image. Thus, the relationship data between the pose key point data and the manual classification result can be obtained.
[0063] S206: Input the pose key point data corresponding to each of the training images into the initial linear regression model to obtain the training classification results corresponding to each of the training images. Then, based on the training classification results and the manual classification results, iteratively train the initial linear regression model until the preset training completion conditions are met and the binary classification model is obtained.
[0064] In this step, the computer equipment can iteratively train the initial linear regression model using training data, including pose keypoint data and manual classification results, until the loss performance of the initial linear regression model meets the preset training completion conditions. If the loss performance of the initial linear regression model meets the preset training completion conditions, it indicates that the initial linear regression model has been trained successfully. This initial linear regression model is a binary classification model that can accurately detect whether an input image is suitable for attribute detection.
[0065] This application obtains a binary classification model through the steps shown in S202 to S206, thereby improving the classification accuracy of the binary classification model and further improving the accuracy of product attribute detection.
[0066] S106: Input each of the product images into the pose detection model to obtain the pose key point data corresponding to each product image.
[0067] In this step, for each product image of the product to be detected, the computer device can input the product image into the posture detection model, so that the posture detection model can detect the posture of the product image and output the posture key point data corresponding to the product image to represent the human posture in the product image.
[0068] S108: Based on the preset image filtering rules and the pose key point data corresponding to each of the product images, select the first target image from each of the product images.
[0069] The image filtering rules refer to the rules used to filter images based on pose key point data. The specific content of these rules can be determined according to the actual situation, and this application does not impose any specific restrictions on them. In this step, the computer device can filter each product image according to the preset image filtering rules and the pose key point data corresponding to each product image, and select the product images that meet the image filtering rules as the first target images.
[0070] In one embodiment, S108 may include: determining whether each product image is a model display image based on the pose key point data corresponding to each product image, and using each model display image as the first target image.
[0071] Specifically, among multiple product images of the item to be detected, there may be various types of images, such as model display images, partial view images, and fabric display images. For product attributes like garment length and sleeve length that cannot be identified from partial view images and fabric display images, the computer equipment can initially filter out model display images from the multiple product images and exclude product images of types such as partial view images and fabric display images. In this way, on the one hand, the number of images processed by the binary classification model can be reduced, improving detection efficiency. On the other hand, it can also eliminate the interference of non-model display images on the binary classification model, making the classification results output by the binary classification model more accurate, thereby further improving detection accuracy.
[0072] In this application, for each product image to be detected, the computer device can determine whether a model exists in the product image based on the pose key point data of the product image output by the pose detection model, and thus determine whether the product image is a model display image. If so, the product image is used as the first target image. In this way, model display images can be selected as the first target images from among the various product images.
[0073] S110: Input the pose key point data corresponding to each of the first target images into the binary classification model to obtain the classification result of each first target image. Each classification result is used to reflect whether the corresponding first target image is suitable for attribute detection.
[0074] Specifically, for each first target image, the computer device can input the pose key point data corresponding to the first target image into a binary classification model to obtain the classification result of the first target image output by the binary classification model, and then determine whether the first target image is suitable for attribute detection based on the classification result of the first target image.
[0075] This application first performs pose detection on the product image, and then performs binary classification based on the pose key point data obtained from the pose detection to determine whether the product image is suitable for attribute detection, thereby improving the accuracy of the binary classification results and thus improving the accuracy of product attribute detection.
[0076] S112: Based on the classification results of each of the first target images, select a second target image suitable for attribute detection from each of the first target images.
[0077] After obtaining the classification results of each first target image, the computer device can determine whether the first target image is suitable for attribute detection based on the classification results of each first target image, and select the first target images suitable for attribute detection as the second target images accordingly.
[0078] S114: Detect the predicted product attribute values of the product to be detected based on each of the second target images.
[0079] In this step, the computer device can obtain the predicted product attributes of the product to be detected based on a second target image suitable for attribute detection from among multiple product images. It is understood that step S114 can be implemented in any manner. In one embodiment, such as... Figure 3 As shown, S114 may include the following steps:
[0080] S302: Determine the attribute to be detected and obtain the single-label multi-classification model corresponding to the attribute to be detected.
[0081] The attribute to be detected refers to the product attribute to be detected, which may include, but is not limited to, clothing length, sleeve length, clothing style, and other product attributes. After determining the attribute to be detected, the computer device can obtain a single-label multi-classification model corresponding to the attribute to be detected, so as to perform attribute detection on the product to be detected through the single-label multi-classification model to obtain the attribute value of the attribute to be detected.
[0082] It is understood that this application can implement a single-label multi-classification model based on any neural network, and can also use any method to train the initial model to obtain a single-label multi-classification model. In one embodiment, during the training of the initial model, this application can use the Focal Loss function as the loss function for training, thereby solving the problem of imbalanced training samples.
[0083] S304: Input each of the second target images into the single-label multi-classification model to obtain the detection attribute values output by the single-label multi-classification model and the confidence level corresponding to each detection attribute value.
[0084] In this model, a single-label multi-class classification model can output at least one detected attribute value and the confidence score corresponding to each detected attribute value based on the acquired input image. For example, the output data of a single-label multi-class classification model may include a first attribute value, the confidence score corresponding to the first attribute value, a second attribute value, and the confidence score corresponding to the second attribute value.
[0085] The computer device can input each second target image into a single-label multi-classification model to perform attribute detection on each second target image through the single-label multi-classification model, and output at least one detected attribute value of the attribute to be detected and the confidence level corresponding to each detected attribute value.
[0086] S306: Based on each of the detected attribute values and the confidence level corresponding to each of the detected attribute values, determine the predicted product attribute value among the detected attribute values.
[0087] For example, computer equipment can use the detection attribute value corresponding to the highest confidence level as the predicted product attribute value of the product to be detected.
[0088] This application implements S114 through S302 to S306, thereby automatically obtaining the predicted product attribute values of the product to be detected through artificial intelligence, which can improve detection efficiency.
[0089] This application introduces a pose detection model and a binary classification model to perform pose detection on each product image. Based on the pose keypoint data output by the pose detection model and the binary classification model, it determines whether each product image is suitable for attribute detection. This allows for the selection of suitable product images from multiple images for attribute detection, and the subsequent attribute detection. This eliminates interference from unsuitable product images, thereby improving the accuracy of attribute detection.
[0090] In one embodiment, such as Figure 4 As shown, the product attribute detection method of this application further includes the following steps:
[0091] S402: Obtain the initial product attribute values of the product to be detected, which are pre-labeled manually;
[0092] S404: If the initial product attribute value is different from the predicted product attribute value, and the confidence level corresponding to the predicted product attribute value is greater than the preset confidence level threshold, then push attribute modification information to the information maintenance personnel.
[0093] Specifically, when uploading product information, suppliers can manually annotate the initial product attribute values for the products to be tested. The computer equipment can compare these manually annotated initial product attribute values with automatically identified predicted product attribute values to determine if correction is needed. If the initial product attribute values differ from the predicted product attribute values, and the confidence level of the predicted product attribute values is greater than a preset confidence threshold, it indicates that the predicted product attribute values are relatively accurate, and the initial product attribute values are highly likely to deviate from the actual situation. In this case, the computer equipment can push attribute modification information to the information maintenance personnel to remind them to modify the product attribute values annotated on the product information. This allows for the checking of product attributes and the pushing of attribute modification information to information maintenance personnel when there is a high probability of error, thereby minimizing the possibility of errors in product information.
[0094] The product attribute detection device provided in the embodiments of this application is described below. The product attribute detection device described below can be referred to in correspondence with the product attribute detection method described above.
[0095] In one embodiment, this application provides a product attribute detection device 500. For example... Figure 5 As shown, the device 500 includes a product image acquisition module 510, a binary classification model acquisition module 520, a pose detection module 530, a first target image filtering module 540, a classification module 550, a second target image filtering module 560, and an attribute value detection module 570. Wherein:
[0096] The product image acquisition module 510 is used to acquire multiple product images of the product to be detected;
[0097] The binary classification model acquisition module 520 is used to acquire the pose detection model and the binary classification model.
[0098] The pose detection module 530 is used to input each of the product images into the pose detection model to obtain the pose key point data corresponding to each product image.
[0099] The first target image filtering module 540 is used to filter out the first target image from each of the product images according to the preset image filtering rules and the posture key point data corresponding to each of the product images;
[0100] The classification module 550 is used to input the pose key point data corresponding to each of the first target images into the binary classification model to obtain the classification result of each first target image. Each classification result is used to reflect whether the corresponding first target image is suitable for attribute detection.
[0101] The second target image filtering module 560 is used to filter out a second target image suitable for attribute detection from each of the first target images based on the classification results of each of the first target images.
[0102] The attribute value detection module 570 is used to determine the predicted product attribute value of the product to be detected based on each of the second target images.
[0103] In one embodiment, the attribute value detection module 570 includes a single-label multi-classification model acquisition unit, a detection unit, and an attribute value determination unit. The single-label multi-classification model acquisition unit is used to determine the attribute to be detected and acquire the single-label multi-classification model corresponding to the attribute to be detected. The detection unit is used to input each of the second target images into the single-label multi-classification model to obtain each detected attribute value output by the single-label multi-classification model and the confidence level corresponding to each detected attribute value. The attribute value determination unit is used to determine the predicted product attribute value among the detected attribute values based on each detected attribute value and the confidence level corresponding to each detected attribute value.
[0104] In one embodiment, the product attribute detection device 500 further includes an attribute value acquisition module and an information push module. The attribute value acquisition module acquires initial product attribute values of the product to be detected, which are pre-labeled manually. The information push module pushes attribute modification information to information maintenance personnel when the initial product attribute values differ from the predicted product attribute values, and the confidence level corresponding to the predicted product attribute values is greater than a preset confidence level threshold.
[0105] In one embodiment, the single-label multi-classification model is trained using the Focal Loss function as the loss function.
[0106] In one embodiment, the first target image filtering module 540 includes a model display image filtering unit. This model display image filtering unit is used to determine whether each product image is a model display image based on the pose keypoint data corresponding to each product image, and to select each model display image as the first target image.
[0107] In one embodiment, the binary classification model acquisition module 520 includes an initial model acquisition unit, a training pose detection unit, and a model training unit. The initial model acquisition unit acquires an initial linear regression model and a training dataset, wherein the training dataset includes multiple training images and pre-labeled manual classification results for each training image. The training pose detection unit inputs each training image into the pose detection model to obtain pose keypoint data corresponding to each training image. The model training unit inputs the pose keypoint data corresponding to each training image into the initial linear regression model to obtain training classification results for each training image, and iteratively trains the initial linear regression model based on the training classification results and the manual classification results until preset training completion conditions are met and the binary classification model is obtained.
[0108] In one embodiment, this application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the commodity attribute detection method described in any of the above embodiments.
[0109] In one embodiment, this application also provides a computer device. The computer device stores computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the product attribute detection method described in any of the above embodiments.
[0110] Indicatively, Figure 6This is a schematic diagram of the internal structure of a computer device provided in an embodiment of this application. In one example, the computer device can be a server. (Refer to...) Figure 6 The computer device 900 includes a processing component 902, which further includes one or more processors, and memory resources represented by memory 901 for storing instructions, such as application programs, that can be executed by the processing component 902. The application programs stored in memory 901 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 902 is configured to execute instructions to perform the steps of the commodity attribute detection method described in any of the above embodiments.
[0111] The computer device 900 may also include a power supply component 903 configured to perform power management of the computer device 900, a wired or wireless network interface 904 configured to connect the computer device 900 to a network, and an input / output (I / O) interface 905. The computer device 900 may operate on an operating system stored in memory 901, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or similar.
[0112] Those skilled in the art will understand that the internal structure of the computer device shown in this application is merely a block diagram of a portion of the structure related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. A specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.
[0113] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this document, "a," "an," "the," "the," and "its" may also include plural forms unless the context clearly indicates otherwise. "Multiple" refers to at least two, such as 2, 3, 5, or 8, etc. "And / or" includes any and all combinations of the related listed items.
[0114] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referred to each other.
[0115] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for detecting product attributes, characterized in that, The method includes: Obtain multiple product images of the product to be inspected; Obtain the pose detection model and the binary classification model; Each of the product images is input into the pose detection model to obtain the pose key point data corresponding to each product image; Based on preset image filtering rules and the pose key point data corresponding to each of the product images, a first target image is selected from each of the product images; specifically, this includes: based on the pose key point data corresponding to each of the product images, determining whether each of the product images is a model display image, and using each of the model display images as the first target image; The pose key point data corresponding to each of the first target images are input into the binary classification model to obtain the classification result of each first target image. Each classification result is used to reflect whether the corresponding first target image is suitable for attribute detection. Based on the classification results of each of the first target images, a second target image suitable for attribute detection is selected from each of the first target images; Based on each of the second target images, the predicted product attribute values of the product to be detected are detected.
2. The product attribute detection method according to claim 1, characterized in that, The step of detecting the predicted product attribute values of the product to be detected based on each of the second target images includes: Determine the attribute to be detected and obtain the single-label multi-classification model corresponding to the attribute to be detected; Each of the second target images is input into the single-label multi-classification model to obtain the detection attribute values output by the single-label multi-classification model and the confidence level corresponding to each detection attribute value; Based on each of the detected attribute values and the confidence level corresponding to each detected attribute value, the predicted product attribute value is determined among the detected attribute values.
3. The product attribute detection method according to claim 2, characterized in that, The method further includes: Obtain the initial product attribute values of the product to be tested, which are pre-labeled by humans; If the initial product attribute value is different from the predicted product attribute value, and the confidence level corresponding to the predicted product attribute value is greater than the preset confidence level threshold, then attribute modification information is pushed to the information maintenance personnel.
4. The product attribute detection method according to claim 2, characterized in that, The single-label multi-classification model was trained using the Focal Loss function as the loss function.
5. The product attribute detection method according to claim 1, characterized in that, The step of selecting the first target image from the product images based on preset image filtering rules and the pose key point data corresponding to each product image includes: Based on the pose key point data corresponding to each product image, it is determined whether each product image is a model display image, and each model display image is used as the first target image.
6. The product attribute detection method according to any one of claims 1 to 5, characterized in that, The steps for obtaining the binary classification model include: Obtain an initial linear regression model and a training dataset, wherein the training dataset includes multiple training images and pre-labeled manual classification results for each training image; Each training image is input into the pose detection model to obtain pose key point data corresponding to each training image; The pose key point data corresponding to each training image are input into the initial linear regression model to obtain the training classification result corresponding to each training image. Based on the training classification result and the manual classification result, the initial linear regression model is iteratively trained until the preset training completion condition is met and the binary classification model is obtained.
7. A product attribute detection device, characterized in that, The device includes: The product image acquisition module is used to acquire multiple product images of the product to be inspected. The binary classification model acquisition module is used to acquire the pose detection model and the binary classification model. The pose detection module is used to input each of the product images into the pose detection model to obtain the pose key point data corresponding to each product image. The first target image filtering module is used to filter out the first target image from each of the product images according to preset image filtering rules and the posture key point data corresponding to each of the product images; specifically, it includes: determining whether each of the product images is a model display image according to the posture key point data corresponding to each of the product images, and using each of the model display images as the first target image; The classification module is used to input the pose key point data corresponding to each of the first target images into the binary classification model to obtain the classification result of each first target image. Each classification result is used to reflect whether the corresponding first target image is suitable for attribute detection. The second target image filtering module is used to filter out second target images suitable for attribute detection from each of the first target images based on the classification results of each first target image. The attribute value detection module is used to determine the predicted product attribute value of the product to be detected based on each of the second target images.
8. The commodity attribute detection device according to claim 7, characterized in that, The attribute value detection module includes: A single-label multi-classification model acquisition unit is used to determine the attribute to be detected and acquire the single-label multi-classification model corresponding to the attribute to be detected. The detection unit is used to input each of the second target images into the single-label multi-classification model to obtain each detection attribute value output by the single-label multi-classification model and the confidence level corresponding to each detection attribute value; An attribute value determination unit is used to determine the predicted product attribute value among the detected attribute values based on each of the detected attribute values and the confidence level corresponding to each of the detected attribute values.
9. A storage medium, characterized in that, The storage medium stores computer-readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the product attribute detection method as described in any one of claims 1 to 6.
10. A computer device, characterized in that, include: One or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, perform the steps of the product attribute detection method as described in any one of claims 1 to 6.