Picture detection method and device, equipment, storage medium and program product
By extracting and fusion images of different resolutions, the computing resource consumption and detection accuracy problems caused by excessive or too small resolution are solved, and efficient multi-task detection is achieved.
Patent Information
- Application Number
- CN202411691703.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-25
- Publication Date
- 2025-02-18
AI Technical Summary
In computer vision technology, excessive resolution of pictures will lead to excessive consumption of computing resources, while too small resolution will affect detection accuracy, making it difficult to effectively extract context and semantic information.
By obtaining pictures of different resolutions of the target picture and inputting them into the feature extraction network of corresponding parameters, a feature map of each resolution is obtained. Then, the feature maps with two resolutions are fused to generate several fusion feature maps, which are finally used for multitask detection.
This method uses feature extraction networks with multiple parameters to save computing resources, and extracts rich context and semantic information, improving the accuracy of multi-task detection.
Smart Images
Figure CN119478444A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of image recognition technology, and in particular to an image detection method, apparatus, computer equipment, storage medium and computer program product. Background Art
[0002] In computer vision technology, AI networks can be used to perform multi-task detection on images, such as target detection, target segmentation, and target classification.
[0003] The resolution of the image as input data has a certain impact on multi-task detection. If the resolution is too high, it will consume more computing resources; if the resolution is too small, the contextual information and semantic information are insufficient, which affects the detection accuracy. Summary of the invention
[0004] Based on this, it is necessary to provide an image detection method, apparatus, computer equipment, storage medium and computer program product to address the above technical problems.
[0005] The present application provides a picture detection method, the method comprising:
[0006] Get images of different resolutions derived from the target image;
[0007] Input the images of each resolution into the feature extraction network with the corresponding parameter amount to obtain the feature map of each resolution; the images with large resolution are input into the feature extraction network with small parameter amount;
[0008] Fusing feature maps of two-to-two resolutions to obtain several fused feature maps;
[0009] According to a plurality of fused feature maps, a multi-task detection result of the target image is obtained.
[0010] In one embodiment, feature maps of two-to-two resolutions are fused to obtain several fused feature maps, including:
[0011] For feature maps of any two resolutions, adding the feature maps of any two resolutions to obtain an added feature map;
[0012] Performing splicing processing on the feature maps of any two resolutions to obtain a spliced feature map;
[0013] According to the added feature map and the concatenated feature map, a fused feature map is obtained.
[0014] In one embodiment, a corresponding fused feature map is obtained according to the added feature map and the concatenated feature map, including:
[0015] Performing convolution processing on the spliced feature map;
[0016] The concatenated feature map and the added feature map after convolution are concatenated to obtain a fused feature map.
[0017] In one embodiment, the concatenated feature map and the added feature map after convolution are concatenated to obtain a fused feature map, including:
[0018] The concatenated feature map and the added feature map after convolution are concatenated and then convolved to obtain a fused feature map.
[0019] In one embodiment, obtaining pictures of different resolutions derived from a target picture includes:
[0020] Get the target image;
[0021] The resolution of the target image is adjusted to obtain images with different resolutions.
[0022] In one embodiment, a multi-task detection result of the target image is obtained according to a plurality of fused feature maps, including:
[0023] Fuse several fusion feature maps to obtain a fusion result;
[0024] According to the fusion result, a multi-task detection result of the target image is obtained.
[0025] The present application provides an image detection device, the device comprising:
[0026] An image acquisition module is used to acquire images of different resolutions from a target image;
[0027] The feature extraction module is used to input images of different resolutions into the feature extraction network with the corresponding parameter amount to obtain feature maps of different resolutions; images with large resolutions are input into the feature extraction network with small parameter amount;
[0028] A fusion processing module is used to fuse feature maps of two resolutions to obtain several fused feature maps;
[0029] The detection result acquisition module is used to obtain the multi-task detection result of the target image according to a plurality of fused feature maps.
[0030] The present application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the above method.
[0031] The present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to execute the above method.
[0032] The present application provides a computer program product, on which a computer program is stored, and the computer program is executed by a processor to perform the above method.
[0033] The above-mentioned image detection method, device, computer equipment, storage medium and computer program product obtain images of different resolutions from the target image; input the images of each resolution into the feature extraction network of the corresponding parameter amount to obtain the feature map of each resolution; the image with a large resolution is input into the feature extraction network with a small parameter amount; the feature maps of the two resolutions are fused to obtain several fused feature maps; based on the several fused feature maps, the multi-task detection result of the target image is obtained. The method provided in the present application introduces a feature extraction network with multiple parameter amounts, inputs the image with a large resolution into the feature extraction network with a small parameter amount, saving computing resources; inputs the image with a small resolution into the feature extraction network with a large parameter amount, which can extract richer contextual information and semantic information to ensure detection accuracy; after obtaining the feature maps of each resolution through the feature extraction network with multiple parameter amounts, the feature maps of the two resolutions are fused to obtain several fused feature maps, and the several fused feature maps are used for multi-task detection to improve the accuracy of multi-task detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0035] Figure 1 Schematic diagram of a process of an image detection method in an embodiment;
[0036] Figure 2 A schematic diagram of the structure of a fusion network in an embodiment;
[0037] Figure 3 is a schematic diagram of the structure of an image detection model in one embodiment;
[0038] Figure 4 is a structural block diagram of a picture detection device in one embodiment;
[0039] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0040] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0041] The image detection method provided in this application can be executed by a computer device, including: Figure 1 Steps shown.
[0042] Step S101, obtaining pictures of different resolutions derived from the target picture.
[0043] The picture to be detected can be called a target picture; after obtaining the target picture, a picture of a first resolution, a picture of a second resolution, and a picture of a third resolution can be obtained based on the target picture; wherein the first resolution, the second resolution, and the third resolution increase in this order, that is, the first resolution is the smallest, the second resolution is in the middle, and the third resolution is the largest.
[0044] Step S102: Input the images of each resolution into the feature extraction network of the corresponding parameter quantity to obtain the feature map of each resolution.
[0045] Images with high resolution are input into a feature extraction network with a small number of parameters; images with low resolution are input into a feature extraction network with a large number of parameters.
[0046] The image detection model includes feature extraction networks with multiple parameter amounts, and the feature extraction network can adopt the structure of the Yolo network. In some scenarios, the image detection model can include feature extraction networks with three parameter amounts. In descending order of parameter amounts, these three feature extraction networks with three parameter amounts are respectively recorded as backbone_L (the feature extraction network with the largest parameter amount), backbone_M (the feature extraction network with a medium parameter amount), and backbone_S (the feature extraction network with the smallest parameter amount).
[0047] After obtaining the picture of the first resolution, the picture of the second resolution and the picture of the third resolution, the first resolution is the smallest. In order to enrich the semantic information and context information, the picture of the first resolution can be input into the feature extraction network backbone_L with the largest parameter amount to obtain the feature map of the first resolution; the second resolution is in the middle, and the picture of the second resolution can be input into the feature extraction network backbone_M with the middle parameter amount to obtain the feature map of the second resolution; the third resolution is the largest. In order to save computing resources, the picture of the third resolution can be input into the feature extraction network backbone_S with the smallest parameter amount to obtain the feature map of the third resolution.
[0048] Step S103, fusing feature maps of two-to-two resolutions to obtain a plurality of fused feature maps;
[0049] Exemplarily, the feature map of the first resolution and the feature map of the second resolution can be fused to obtain a fused feature map; the feature map of the first resolution and the feature map of the third resolution can be fused to obtain a fused feature map; the feature map of the second resolution and the feature map of the third resolution can be fused to obtain a fused feature map.
[0050] In some scenarios, the image detection model includes several fusion networks; the feature map of the first resolution and the feature map of the second resolution can be input into the fusion network to obtain the corresponding fusion feature map; the feature map of the first resolution and the feature map of the third resolution can be input into another fusion network to obtain the corresponding fusion feature map; the feature map of the second resolution and the feature map of the third resolution can be input into another fusion network to obtain the corresponding fusion feature map.
[0051] Step S104, obtaining a multi-task detection result of the target image according to a plurality of fused feature maps.
[0052] In some scenarios, the image detection model also includes multiple types of task detection networks; inputting several fused feature maps into a certain type of task detection network can obtain the corresponding type of detection results; inputting several fused feature maps into another type of task detection network can obtain the corresponding type of detection results. The types of tasks include object detection, object segmentation, and object classification.
[0053] In the above-mentioned image detection method, images of different resolutions are obtained from the target image; the images of each resolution are input into the feature extraction network of the corresponding parameter amount to obtain the feature map of each resolution; the image with a large resolution is input into the feature extraction network with a small parameter amount; the feature maps of the two resolutions are fused to obtain several fused feature maps; based on the several fused feature maps, the multi-task detection result of the target image is obtained. The method provided in the present application introduces a feature extraction network with multiple parameter amounts, and the image with a large resolution is input into the feature extraction network with a small parameter amount to save computing resources; the image with a small resolution is input into the feature extraction network with a large parameter amount, so that richer contextual information and semantic information can be extracted to ensure detection accuracy; after obtaining the feature maps of each resolution through the feature extraction network with multiple parameter amounts, the feature maps of the two resolutions are fused to obtain several fused feature maps, and the several fused feature maps are used for multi-task detection to improve the accuracy of multi-task detection.
[0054] In one embodiment, feature maps of two-to-two resolutions are fused to obtain several fused feature maps, including:
[0055] For feature maps of any two resolutions, the feature maps of any two resolutions are added to obtain an added feature map; the feature maps of any two resolutions are spliced to obtain a spliced feature map; and a fused feature map is obtained based on the added feature map and the spliced feature map.
[0056] Exemplarily, for a feature map of a first resolution and a feature map of a second resolution, the feature map of the first resolution and the feature map of the second resolution can be added to obtain an added feature map, which corresponds to the first resolution and the second resolution; the feature map of the first resolution and the feature map of the second resolution can be spliced to obtain a spliced feature map, which corresponds to the first resolution and the second resolution; based on the added feature map and the spliced feature map, a fused feature map can be obtained, which corresponds to the first resolution and the second resolution.
[0057] As another example, for the feature map of the first resolution and the feature map of the third resolution, the feature map of the first resolution and the feature map of the third resolution can be added to obtain an added feature map, which corresponds to the first resolution and the third resolution; the feature map of the first resolution and the feature map of the third resolution can be spliced to obtain a spliced feature map, which corresponds to the first resolution and the third resolution; based on the added feature map and the spliced feature map, a fused feature map can be obtained, which corresponds to the first resolution and the third resolution.
[0058] In one embodiment, a corresponding fused feature map is obtained according to the added feature map and the concatenated feature map, including:
[0059] The concatenated feature map is convolved; the convolved concatenated feature map and the added feature map are concatenated to obtain a fused feature map.
[0060] Exemplarily, for a feature map of a first resolution and a feature map of a second resolution, after obtaining an added feature map and a spliced feature map corresponding to the first resolution and the second resolution, the spliced feature map can be convolved; the spliced feature map after the convolution is spliced with the added feature map to obtain a fused feature map, which corresponds to the first resolution and the second resolution.
[0061] As another example, for the feature map of the first resolution and the feature map of the third resolution, after obtaining the added feature map and the spliced feature map corresponding to the first resolution and the third resolution, the spliced feature map can be convolved; the convolved spliced feature map and the added feature map are spliced to obtain a fused feature map, which corresponds to the first resolution and the third resolution.
[0062] In one embodiment, the concatenated feature map and the added feature map after convolution are concatenated to obtain a fused feature map, including:
[0063] The concatenated feature map and the added feature map after convolution are concatenated and then convolved to obtain a fused feature map.
[0064] Exemplarily, for a feature map of a first resolution and a feature map of a second resolution, after obtaining an added feature map and a spliced feature map corresponding to the first resolution and the second resolution, the spliced feature map can be convolved; after splicing the convolved spliced feature map and the added feature map, convolution is performed again to obtain a fused feature map, which corresponds to the first resolution and the second resolution.
[0065] As another example, for the feature map of the first resolution and the feature map of the third resolution, after obtaining the added feature map and the spliced feature map corresponding to the first resolution and the third resolution, the spliced feature map can be convolved; after splicing the convolved spliced feature map and the added feature map, convolution is performed again to obtain a fused feature map, which corresponds to the first resolution and the third resolution.
[0066] In some scenarios, the image detection model includes several fusion networks, each of which includes two branches, one branch has an addition unit, and the other branch has a first splicing unit and a first convolution unit connected in sequence; the output results of the two branches are summarized in the second splicing unit, and the second convolution unit is connected after the second splicing unit, such as Figure 2 shown.
[0067] Exemplarily, for a feature map of a first resolution and a feature map of a second resolution, the feature map of the first resolution and the feature map of the second resolution can be input into one of the fusion networks. The feature map of the first resolution and the feature map of the second resolution are added by the addition unit of the fusion network to obtain an added feature map; the feature dimensions of the feature map of the first resolution and the feature map of the second resolution are [N, C, H, W], and the feature dimensions of the added feature map are [N, C, H, W]; the feature map of the first resolution and the feature map of the second resolution are spliced in the feature dimension by the first splicing unit of the fusion network; wherein the feature dimension of the spliced feature map output by the first splicing unit is [N, 2×C, W, H]. wherein N represents the number of images, C represents the number of channels of the feature map, H represents the height of the feature map, and W represents the width of the feature map.
[0068] The spliced feature map output by the first splicing unit is input into the first convolution unit to obtain the spliced feature map after convolution. The feature dimension of the spliced feature map after convolution is [N, C, W, H].
[0069] The added feature map and the convolved feature map are input into the second concatenation unit for concatenation, and the feature dimension of the result output by the second concatenation unit is [N, 2×C, W, H]; the result output by the second concatenation unit is further convolved by the second convolution unit to obtain a fused feature map, and the feature dimension of the fused feature map is [N, C, W, H].
[0070] In one embodiment, obtaining pictures of different resolutions derived from a target picture includes:
[0071] Get the target image; adjust the resolution of the target image to obtain images with different resolutions.
[0072] The image to be detected can be used as the target image. For example, the resolution of the target image can be adjusted to a first resolution to obtain a first resolution image; the resolution of the target image can be adjusted to a second resolution to obtain a second resolution image; the resolution of the target image can be adjusted to a third resolution to obtain a third resolution image.
[0073] In one embodiment, a multi-task detection result of a target image is obtained based on a plurality of fused feature maps, including:
[0074] Several fusion feature maps are fused to obtain a fusion result; based on the fusion result, a multi-task detection result of the target image is obtained.
[0075] In this embodiment, in order to reduce the complexity of multi-task detection, after obtaining several fused feature maps, the several fused feature maps can be fused to obtain a fusion result; multi-task detection is performed based on the fusion result to obtain a multi-task detection result of the target image; specifically, the fusion result can be input into multiple types of task detection networks respectively to obtain multiple types of detection results.
[0076] In order to better understand the above method, an application example of the image detection method of the present application is described in detail below. Figure 3 ,The image detection model includes three parameter quantity feature extraction networks, three fusion networks and a fusion unit.
[0077] During the training phase, different tasks can be trained based on different training data sets. For multiple tasks, the loss values of each task can be merged to obtain a combined loss value, and the model parameters can be updated based on the combined loss value. If the number of tasks is increased, a larger network can be used to improve network generalization.
[0078] In the inference stage, after obtaining the pictures of the first resolution, the second resolution and the third resolution, the first resolution is the smallest. In order to enrich the semantic information and context information, the picture of the first resolution can be input into the feature extraction network backbone_L with the largest parameter amount to obtain the feature map of the first resolution; the second resolution is in the middle, and the picture of the second resolution can be input into the feature extraction network backbone_M with the middle parameter amount to obtain the feature map of the second resolution; the third resolution is the largest. In order to save computing resources, the picture of the third resolution can be input into the feature extraction network backbone_S with the smallest parameter amount to obtain the feature map of the third resolution.
[0079] The feature map of the first resolution and the feature map of the second resolution are input into the first fusion network to obtain a fused feature map. The feature map of the first resolution and the feature map of the third resolution are input into the second fusion network to obtain a fused feature map. The feature map of the second resolution and the feature map of the third resolution are input into the third fusion network to obtain a fused feature map.
[0080] The fusion feature maps output by the first fusion network, the second fusion network and the third fusion network are respectively input into the fusion unit to obtain a fusion result.
[0081] The fusion results are input into multiple types of task detection networks respectively to obtain multiple types of detection results.
[0082] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0083] Based on the same inventive concept, the embodiment of the present application also provides a picture detection device for implementing the picture detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more picture detection device embodiments provided below can refer to the limitations on the picture detection method above, and will not be repeated here.
[0084] In one embodiment, Figure 4 As shown, a picture detection device is provided, comprising:
[0085] The picture acquisition module 401 is used to acquire pictures of different resolutions from the target picture;
[0086] The feature extraction module 402 is used to input the pictures of each resolution into the feature extraction network of the corresponding parameter amount to obtain the feature map of each resolution; the pictures with large resolution are input into the feature extraction network with small parameter amount;
[0087] A fusion processing module 403 is used to perform fusion processing on feature maps of two-to-two resolutions to obtain a plurality of fused feature maps;
[0088] The detection result acquisition module 404 is used to obtain the multi-task detection result of the target image according to a plurality of fused feature maps.
[0089] In one embodiment, the fusion processing module 403 is further used to:
[0090] For feature maps of any two resolutions, the feature maps of any two resolutions are added to obtain an added feature map; the feature maps of any two resolutions are spliced to obtain a spliced feature map; and a fused feature map is obtained based on the added feature map and the spliced feature map.
[0091] In one embodiment, the fusion processing module 403 is further used to:
[0092] The concatenated feature map is subjected to convolution processing; the concatenated feature map and the added feature map after convolution are subjected to concatenation processing to obtain a fused feature map.
[0093] In one embodiment, the fusion processing module 403 is further used to:
[0094] The concatenated feature map and the added feature map after convolution are concatenated and then convolved to obtain a fused feature map.
[0095] In one embodiment, the picture acquisition module 401 is further used to:
[0096] Acquire a target image; and adjust the resolution of the target image to obtain images with different resolutions.
[0097] In one embodiment, the detection result acquisition module 404 is further used to:
[0098] A plurality of fusion feature maps are fused to obtain a fusion result; and a multi-task detection result of the target image is obtained according to the fusion result.
[0099] Each module in the above-mentioned image detection device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.
[0100] In an exemplary embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data involved in the above method. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a picture detection method is implemented.
[0101] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0102] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned various method embodiments when executing the computer program.
[0103] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.
[0104] In one embodiment, a computer program product is provided, on which a computer program is stored, and the computer program is used by a processor to execute the steps in the above-mentioned various method embodiments.
[0105] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0106] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.
[0107] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0108] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A picture detection method, characterized in that: The method comprises: Get images of different resolutions derived from the target image; Input the images of each resolution into the feature extraction network with the corresponding parameter amount to obtain the feature map of each resolution; the images with large resolution are input into the feature extraction network with small parameter amount; Fusing feature maps of two-to-two resolutions to obtain several fused feature maps; According to a plurality of fused feature maps, a multi-task detection result of the target image is obtained.
2. The method according to claim 1, characterized in that The feature maps of each resolution are fused to obtain several fused feature maps, including: For feature maps of any two resolutions, adding the feature maps of any two resolutions to obtain an added feature map; Performing splicing processing on the feature maps of any two resolutions to obtain a spliced feature map; According to the added feature map and the concatenated feature map, a fused feature map is obtained.
3. The method according to claim 2, characterized in that According to the added feature map and the concatenated feature map, the corresponding fused feature map is obtained, including: Performing convolution processing on the spliced feature map; The concatenated feature map and the added feature map after convolution are concatenated to obtain a fused feature map.
4. The method according to claim 3, characterized in that The concatenated feature map and the added feature map after convolution are concatenated to obtain a fused feature map, including: The concatenated feature map and the added feature map after convolution are concatenated and then convolved to obtain a fused feature map.
5. The method according to claim 1, characterized in that Get images of different resolutions from the target image, including: Get the target image; The resolution of the target image is adjusted to obtain images with different resolutions.
6. The method according to any one of claims 1 to 5, characterized in that: According to several fused feature maps, a multi-task detection result of the target image is obtained, including: Fuse several fusion feature maps to obtain a fusion result; According to the fusion result, a multi-task detection result of the target image is obtained.
7. A picture detection device, characterized in that: The device comprises: An image acquisition module is used to acquire images of different resolutions from a target image; The feature extraction module is used to input the images of each resolution into the feature extraction network of the corresponding parameter amount to obtain the feature map of each resolution; the images with large resolution are input into the feature extraction network with small parameter amount; A fusion processing module is used to fuse feature maps of two resolutions to obtain several fused feature maps; The detection result acquisition module is used to obtain the multi-task detection result of the target image according to a plurality of fused feature maps.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.