Agricultural product video generation method and device, electronic equipment and storage medium
The method of reconstructing migration images by estimating human pose and depth solves the problem of users being unable to efficiently produce agricultural product videos, thus achieving efficient video production and increased revenue for merchants.
Patent Information
- Application Number
- CN202511729509.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-24
- Publication Date
- 2026-02-17
AI Technical Summary
Existing technologies prevent users from efficiently creating videos related to agricultural products, as they involve high costs and difficulties in designing and recording video content.
By acquiring user-selected sample videos and original images, human pose estimation and depth estimation are performed, the transformation matrix is calculated, the migration images are reconstructed using a generative network, and then stitched together to form the target video.
It enabled the efficient production of agricultural product videos, improved product recommendation effectiveness, increased merchant revenue, and enhanced the platform's attractiveness.
Smart Images
Figure CN121547640A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, and in particular to an agricultural product video generation method and device, an electronic device, and a storage medium. BACKGROUND
[0002] Agricultural product recommendation videos have great significance for increasing product sales, increasing merchant income, and improving the lives of merchants. However, there are currently problems such as high cost, difficult video content design, and difficult video recording in video production, which all result in users being unable to efficiently produce agricultural product-related videos. SUMMARY
[0003] Therefore, embodiments of the present application provide an agricultural product video generation method and device, an electronic device, and a storage medium to solve the problem that users are unable to efficiently produce agricultural product-related videos in the prior art.
[0004] To achieve the above object, embodiments of the present application provide the following technical solutions.
[0005] The first aspect shows an agricultural product video generation method, which comprises:
[0006] obtaining a sample video selected by a user and an original image uploaded by the user;
[0007] determining a first human body pose corresponding to the original image and a second human body pose corresponding to each sample image in the sample video based on the original image and each sample image in the sample video;
[0008] processing the original image, the first human body pose, and the second human body pose corresponding to each sample image to determine a reconstructed migration image of each frame;
[0009] splicing the reconstructed migration image of each frame to obtain a target video.
[0010] Optionally, determining the first human body pose corresponding to the original image and the second human body pose corresponding to each sample image based on the original image and each sample image in the sample video comprises:
[0011] frame processing the sample video to obtain a plurality of sample images;
[0012] respectively performing depth estimation on the original image and each sample image to obtain a first depth map corresponding to the original image and a second depth map corresponding to each sample image;
[0013] respectively performing human body pose estimation on the first depth map and the second depth map to obtain the first human body pose corresponding to the original image and the second human body pose corresponding to each sample image.
[0014] Optionally, the processing is performed according to the original image, the first human body pose, and the second human body pose corresponding to each frame of sample image, to determine a reconstructed migration image of each frame, including:
[0015] calculating a conversion matrix according to the first human body pose and each second human body pose;
[0016] converting the original image into a migration image of each frame based on the conversion matrix;
[0017] for the migration image of each frame, performing image reconstruction based on the migration image, the original image, the second depth image, and the second human body pose to obtain a plurality of reconstructed migration images.
[0018] Optionally, for the migration image of each frame, the image reconstruction is performed based on the migration image, the original image, the second depth image, and the second human body pose to obtain a reconstructed migration image, including:
[0019] dividing the original image into a background image and a foreground image;
[0020] processing the background image by using an obtained background generation network to obtain a reconstructed background image;
[0021] for the migration image of each frame, performing pose migration image generation on the migration image, the second depth image, and the second human body pose by using an obtained image generation network to obtain a pose migration foreground image;
[0022] integrating the pose migration foreground image and the reconstructed background image to obtain a reconstructed migration image.
[0023] Optionally, the method further includes:
[0024] performing image reconstruction based on the original image, the first depth image, and the first human body pose to obtain a reconstructed original image;
[0025] updating the image generation network based on a similarity between the reconstructed original image and the original image.
[0026] Optionally, the image reconstruction based on the original image, the first depth image, and the first human body pose to obtain a reconstructed original image includes:
[0027] dividing the original image into a background image and a foreground image;
[0028] processing the background image by using an obtained background generation network to obtain a reconstructed background image;
[0029] performing image generation on the foreground image, the first depth image, and the first human body pose by using an obtained image generation network to obtain a first foreground image.
[0030] Based on the integration of the first foreground image and the reconstructed background image, a reconstructed original image is obtained.
[0031] The second aspect shows an agricultural product video generation device, which comprises:
[0032] An acquisition unit is configured to acquire a sample video selected by a user and an uploaded original image.
[0033] A human pose generation unit is configured to determine a first human pose corresponding to the original image and a second human pose corresponding to each frame of sample image based on the original image and each frame of sample image in the sample video.
[0034] A processing unit is configured to process the original image, the first human pose and the second human pose corresponding to each frame of sample image to determine a reconstructed migration image of each frame.
[0035] A video generation unit is configured to splice the reconstructed migration image of each frame to obtain a target video.
[0036] Optionally, the human pose generation unit is specifically configured to:
[0037] frame processing is performed on the sample video to obtain a plurality of sample images;
[0038] depth estimation is performed on the original image and each frame of sample image respectively to obtain a first depth map corresponding to the original image and a second depth map corresponding to each frame of sample image;
[0039] human pose estimation is performed on the first depth map and the second depth map respectively to obtain the first human pose corresponding to the original image and the second human pose corresponding to each frame of sample image.
[0040] The third aspect shows an electronic device, which comprises a processor and a memory, the memory is configured to store data generated program code and data, and the processor is configured to call program instructions in the memory to execute the agricultural product video method as described in any one of the first aspect.
[0041] The fourth aspect shows a storage medium, which comprises a program, wherein when the program runs, the device where the storage medium is located is controlled to execute the agricultural product video method as described in any one of the first aspect.
[0042] Based on the agricultural product video generation method, device, electronic equipment and storage medium provided by the embodiment of the present application, the method comprises: obtaining a sample video selected by a user and an uploaded original image; determining a first human body posture corresponding to the original image and a second human body posture corresponding to each sample image in the sample video based on the original image and each sample image in the sample video; processing the original image, the first human body posture and the second human body posture corresponding to each sample image to determine a reconstructed migration image of each frame; and splicing the reconstructed migration image of each frame to obtain a target video. The present application determines the reconstructed migration image of each frame through human body posture reconstruction, posture synthesis, depth texture processing and the like of the original image and each sample image in the sample video, and splices the reconstructed migration image to obtain the target video. Thus, the production of the agricultural product video is efficiently realized, and the commodity recommendation, the attraction of the guest flow capacity of the platform and the increase of the income of the merchant are realized. BRIEF DESCRIPTION OF DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor based on the provided drawings.
[0044] Figure 1 The flowchart of the agricultural product video generation method shown by the embodiment of the present application is shown.
[0045] Figure 2 The human body posture reconstruction flowchart shown by the embodiment of the present application is shown.
[0046] Figure 3 The flowchart of the reconstructed migration image shown by the embodiment of the present application is shown.
[0047] Figure 4 The flowchart of the reconstructed migration image generated based on the original image and the sample image shown by the embodiment of the present application is shown.
[0048] Figure 5 The structural diagram of the agricultural product video generation device shown by the embodiment of the present application is shown. DETAILED DESCRIPTION
[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.
[0050] The terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of these terms herein is to be construed to cover a changeable order, sequence or arrangement, for example, unless otherwise stated or understood by a person of ordinary skill in the art. Furthermore, the terms "comprising", "including", "containing", and any other similar words, are intended to be equivalent to each other in indicating the inclusion of a stated item, step, process, act, or the like, but not to the exclusion of any additional item, step, process, act, or the like.
[0051] It should be noted that the terms "first", "second", "third", "fourth" and the like in the description and in the claims of the present application, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the use of these terms herein is to be construed to cover a changeable order, sequence or arrangement, for example, unless otherwise stated or understood by a person of ordinary skill in the art. Furthermore, the terms "comprising", "including", "containing", and any other similar words, are intended to be equivalent to each other in indicating the inclusion of a stated item, step, process, act, or the like, but not to the exclusion of any additional item, step, process, act, or the like.
[0052] In this application, the terms "comprising", "containing" or any other similar words are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. Without more limitations, the element defined by the sentence "including a…" does not exclude the presence of other identical elements in the process, method, article or equipment including the said element.
[0053] Referring to Figure 1 A flowchart of an agricultural product video generation method according to an embodiment of the present application is shown in the figure, which comprises the following steps:
[0054] Step S101: Obtain the sample video selected by the user and the original image uploaded by the user;
[0055] In the process of implementing step S101, the server responds to the sample video selected by the user for the target application in the client, and the original image uploaded by the client.
[0056] It should be noted that when the target application receives the video generation instruction, different types of sample videos are displayed to the user for selection.
[0057] It should be noted that the original image, i.e. the original image, is a person image.
[0058] Step S102: determining a first human pose corresponding to the original image and a second human pose corresponding to each frame of the sample image based on the original image and each frame of the sample image in the sample video;
[0059] It should be noted that in the process of implementing step S102, the following steps are included.
[0060] Step S11: frame processing of the sample video to obtain multiple frames of sample images;
[0061] In the method for implementing step S11, the sample video is divided to obtain multiple frames of sample images.
[0062] Step S12: depth estimation is performed on the original image and each frame of the sample image respectively to obtain a first depth map corresponding to the original image and a second depth map corresponding to each frame of the sample image;
[0063] In the process of implementing step S12, for the original image, it is input into a preset depth estimation network, and its corresponding depth probability distribution is output; then the depth probability distribution is weighted and averaged with the depth interval distribution value to obtain the corresponding depth map, which is taken as the first depth map. Similarly, each frame of the sample image is processed in the above manner to obtain the second depth map.
[0064] The depth estimation network can reduce depth estimation error and accelerate convergence, and the obtained full image depth map is supervised by using the label full image depth map in the training stage.
[0065] It should be noted that the depth estimation network is a depth estimation network with the structure of a depth learning model u-net.
[0066] Step S13: human pose estimation is performed on the first depth map and the second depth map respectively to obtain a first human pose corresponding to the original image and a second human pose corresponding to each frame of the sample image.
[0067] In the specific implementation step S13, for the first depth map, the first depth map and the initial image are concatenated and input into the pre-constructed human pose reconstruction network for processing, and the first human pose corresponding to the original image is output; for each second depth map, the second depth map and the sample image of that frame are concatenated and input into the pre-constructed human pose reconstruction network for processing, and the second human pose corresponding to each sample image is output; that is, pose recovery is performed by combining depth maps, and more accurate human pose recovery is promoted by introducing explicit depth information.
[0068] It should be noted that the first and second human poses refer to colored 3D meshes. During the training phase, labeled 3D geometric meshes (human poses) are used.
[0069] It should be noted that the specific implementation steps S11 to S13 illustrate the process of human posture reconstruction, such as... Figure 2 As shown.
[0070] The pre-built human pose reconstruction network is pre-built based on the depth of multiple historical images. Figure 1 The machine is trained to correspond to human postures.
[0071] A pre-built human pose reconstruction network can be used for a probabilistic voting aggregation algorithm for full-map depth estimation.
[0072] The first depth map is the original image depth map, and the second depth map is the sample image depth map.
[0073] Step S103: Process the original image, the first human pose, and the second human pose corresponding to each sample image to determine the reconstructed migration image of each frame;
[0074] It should be noted that the specific implementation of step S103 includes the following steps.
[0075] Step S21: Calculate the transformation matrix based on the first human pose and each second human pose;
[0076] In the specific implementation of step S21, for the second human pose corresponding to each frame of sample image, the second human pose is used as the second vector and the first human pose is used as the first vector; the first vector and the second vector are mapped to obtain the transformation matrix.
[0077] Step S22: Convert the original image into a migration map for each frame based on the transformation matrix;
[0078] In the specific implementation step S22, for each sample image, the original image is converted into a coarse transfer map through the transformation matrix, that is, only the action is transferred. At this time, the texture details are not faithful enough and there is a relatively serious distortion. Therefore, it is necessary to process it with the generative adversarial network in the following text.
[0079] Step S23: For each frame of the migration map, perform image reconstruction based on the migration map, the original image, the second depth map, and the second human pose to obtain multi-frame reconstructed migration images.
[0080] It should be noted that the specific implementation of step S23 includes the following steps.
[0081] Step S31: Divide the original image into a background image and a foreground image;
[0082] In the specific implementation step S31, the original image is simply divided into a foreground image and a background image by using the first human pose. That is, the human pose of the first human pose roughly outlines the area where the foreground is located in the original image, and the original image is simply divided into a foreground image and a background image based on this.
[0083] The foreground image is a person.
[0084] It should be noted that the background image here contains the outline of the foreground, so a background generation network is required for processing.
[0085] Step S32: Process the background image using the acquired background generation network to obtain a reconstructed background image;
[0086] In the specific implementation of step S32, the background image, i.e. the incomplete background image, is completed using the acquired background generation network to obtain the complete background image, i.e. the reconstructed background image.
[0087] It should be noted that the background generation network is pre-trained based on different incomplete background images and their corresponding complete background images; it is an adversarial network.
[0088] Step S33: For each frame of the migration map, use the acquired graph generation network to generate a pose migration image from the migration map, the second depth map, and the second human pose, to obtain a pose migration foreground image.
[0089] It should be noted that graph generation networks include original graph generation networks and transition graph generation networks, both of which are adversarial networks.
[0090] Optionally, the following steps may be included before performing step S33.
[0091] Step S41: Reconstruct the image based on the original image, the first depth map, and the first human pose to obtain the reconstructed original image;
[0092] It should be noted that the specific implementation process of step S41 includes the following steps.
[0093] Step S51: Divide the original image into a background image and a foreground image;
[0094] Step S52: Process the background image using the acquired background generation network to obtain a reconstructed background image;
[0095] It should be noted that the specific implementation processes of steps S51 and S52 are the same as those of steps S31 and S32 mentioned above, and can be referred to each other.
[0096] Step S53: Use the acquired graph generation network to generate images from the foreground image, the first depth map, and the first human pose to obtain the first foreground image;
[0097] In the specific implementation step S53, the foreground image, the first depth map and the first human pose are concatenated and input into the original image generation network for processing, so as to output a complete foreground image, namely the first foreground image.
[0098] It should be noted that the purpose of the original image generation network is to leverage the generative adversarial network's ability to effectively recover the texture information of the original image and then transfer information to the final transfer image generative adversarial network.
[0099] Step S54: Based on the first foreground image and the reconstructed background image, the reconstructed original image is obtained.
[0100] In the specific implementation step S54, the reconstructed background image is added to the first foreground image to obtain the complete original image, that is, the reconstructed original image.
[0101] Step S42: Update the graph generation network based on the similarity between the reconstructed original image and the original image.
[0102] In one implementation, the similarity between the reconstructed original image and the original image is first calculated. If the similarity is greater than or equal to a preset threshold, the loss function of the graph generation network is calculated based on the similarity between the reconstructed original image and the original image; otherwise, the original image is reconstructed. The original image generation network is then updated based on the loss function.
[0103] In another embodiment, the similarity between the reconstructed original image and the original image is calculated. If the similarity is greater than or equal to a preset threshold, it is determined that the original image generation network can transmit information to the migration graph generation adversarial network. If the similarity is less than the preset threshold, the original image is reconstructed, that is, the original image generation network is updated.
[0104] In the specific implementation of step S33, for each frame of the migration map, the migration map generation network in the graph generation network and the original image generation network processed by steps S41 to S42 are obtained; the migration map, the second depth map and the second human pose are used as inputs, and the image after pose transfer is generated by the migration map generation network. In this process, the features generated by each layer of the migration map generation network are summed with the features of the corresponding layer in the original image generation network to obtain the output of the next layer of the migration map generation network, that is, the reconstructed migration image, thus obtaining the reconstructed migration image of each frame.
[0105] Among them, the depth features of the second depth map contain the true distribution of texture features in space, which can better help the transfer map generation network to recover the texture of people in the transferred image.
[0106] In the process of reconstructing the transferred image, this application combines the transfer image generation network and the original image generation network processed by steps S41 to S42. This not only restores the human posture, but also preserves the texture features of the person in the original image. This processing method improves the texture of the person in the generated image, making it more authentic.
[0107] Step S34: Integrate the pose transfer foreground image with the reconstructed background image to obtain the reconstructed transfer image.
[0108] In the specific implementation step S34, the reconstructed background image is added to the pose transfer foreground image to obtain a complete image, namely the reconstructed transfer image.
[0109] It should be noted that the specific implementation step S103 illustrates the process of reconstructing the migrated image, such as... Figure 3 As shown.
[0110] Optionally, based on steps S101 to S103, this application performs probability voting aggregation of the original image and each sample image to obtain the original image depth map (i.e., the first depth map) and the sample image depth map (i.e., the second depth map); then, it performs human pose estimation on the original image and the first depth map to obtain the first human pose; it also performs human pose estimation on the sample image and the second depth map to obtain the second human pose; then, it calculates the transformation matrix between the vectors of the first and second human poses; and based on the transformation matrix, the background and foreground images divided from the original image, the first depth map, and / or the second depth map, etc., it passes through a background generation network, an original image generation network, and a migration map generation network to determine the reconstructed migration image for each frame, such as... Figure 4 As shown.
[0111] Step S104: Stitch together the reconstructed migration images of each frame to obtain the target video.
[0112] In the specific implementation step S104, the reconstructed migration images are stitched together according to the order of each frame during the division to obtain the target video.
[0113] In this embodiment of the invention, the original image of a person input by the user can be transferred to each frame of a sample video through human pose generation, pose synthesis, and processing of reconstructed transfer images. This generates a reconstructed transfer image that is identical to the original image's person, and has the same actions and background as the sample images. By repeating the above process for each frame of the sample image, a series of reconstructed transfer images can be obtained. These images are then stitched together to obtain the final marketing video, i.e., the target video. This efficiently enables the production of agricultural product videos, thereby facilitating product recommendations, improving the platform's ability to attract customers, and increasing merchants' revenue.
[0114] Based on the agricultural product video generation device shown in the above embodiments of the present invention, such as Figure 5 As shown, the device includes:
[0115] The acquisition unit 501 is used to acquire the sample video selected by the user and the uploaded original image;
[0116] The human pose generation unit 502 is used to determine the first human pose corresponding to the original image and the second human pose corresponding to each frame of the sample video based on the original image and each frame of the sample video.
[0117] The processing unit 503 is used to process the original image, the first human pose and the second human pose corresponding to each frame of sample image to determine the reconstructed migration image of each frame.
[0118] The video generation unit 504 is used to stitch together the reconstructed migration images of each frame to obtain the target video.
[0119] The specific principles and execution processes of each unit in the agricultural product video device disclosed in the above embodiments of the present invention are the same as the corresponding contents in the agricultural product video method provided in the above embodiments of the present invention. Please refer to the corresponding parts in the agricultural product video method disclosed in the above embodiments of the present invention, and they will not be repeated here.
[0120] In this embodiment of the invention, the original image of a person input by the user can be transferred to each frame of a sample video through human pose generation, pose synthesis, and processing of reconstructed transfer images. This generates a reconstructed transfer image that is identical to the original image's person, and has the same actions and background as the sample images. By repeating the above process for each frame of the sample image, a series of reconstructed transfer images can be obtained. These images are then stitched together to obtain the final marketing video, i.e., the target video. This efficiently enables the production of agricultural product videos, thereby facilitating product recommendations, improving the platform's ability to attract customers, and increasing merchants' revenue.
[0121] Optionally, based on the agricultural product video generation device shown in the above embodiments of the present invention, the human posture generation unit 502 is specifically used for:
[0122] The sample video is divided into frames to obtain multiple sample images;
[0123] Depth estimation is performed on the original image and each of the sample images to obtain a first depth map corresponding to the original image and a second depth map corresponding to each of the sample images.
[0124] Human pose estimation is performed on the first depth map and the second depth map respectively to obtain the first human pose corresponding to the original image and the second human pose corresponding to each sample image frame.
[0125] Optionally, based on the agricultural product video generation apparatus shown in the above embodiments of the present invention, the processing unit 503 is specifically used for:
[0126] Calculate the transformation matrix based on the first human posture and each second human posture;
[0127] Based on the transformation matrix, the original image is converted into a migration map for each frame;
[0128] For each frame of the migration map, image reconstruction is performed based on the migration map, the original image, the second depth map, and the second human pose to obtain multiple frames of reconstructed migration images.
[0129] Specifically, for each frame's migration map, image reconstruction is performed based on the migration map, the original image, the second depth map, and the second human pose to obtain a reconstructed migration image, including:
[0130] The original image is divided into a background image and a foreground image;
[0131] The background image is processed using the acquired background generation network to obtain a reconstructed background image;
[0132] For each frame of the migration map, the acquired graph generation network is used to generate a pose migration image from the migration map, the second depth map, and the second human pose, thus obtaining a pose migration foreground image.
[0133] The reconstructed migration image is obtained by integrating the pose migration foreground image with the reconstructed background image.
[0134] Optionally, based on the agricultural product video generation apparatus shown in the above embodiments of the present invention, the processing unit 503 is further configured to:
[0135] Image reconstruction is performed based on the original image, the first depth map, and the first human pose to obtain the reconstructed original image;
[0136] The graph generation network is updated based on the similarity between the reconstructed original image and the original image.
[0137] The process of image reconstruction based on the original image, the first depth map, and the first human pose to obtain the reconstructed original image includes:
[0138] The original image is divided into a background image and a foreground image;
[0139] The background image is processed using the acquired background generation network to obtain a reconstructed background image;
[0140] The obtained graph generation network is used to generate an image from the foreground image, the first depth map, and the first human pose to obtain the first foreground image;
[0141] The reconstructed original image is obtained by integrating the first foreground image with the reconstructed background image.
[0142] This application provides an electronic device, which includes a processor and a memory. The memory is used to store agricultural product video program code and data, and the processor is used to call the program instructions in the memory to execute the steps shown in the agricultural product video method in the above embodiments.
[0143] This invention provides a storage medium, namely a computer-readable storage medium, which includes the electronic device provided in the above-described embodiments of this application. The electronic device is used to execute the agricultural product video method disclosed in the embodiments of this application.
[0144] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.
[0145] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.
[0146] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. 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 the invention. Therefore, the invention 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 generating videos of agricultural products, characterized in that, The method includes: Obtain the sample video selected by the user and the uploaded original image; Based on the original image and each frame of the sample video, determine the first human pose corresponding to the original image and the second human pose corresponding to each frame of the sample video. The original image, the first human pose, and the second human pose corresponding to each sample image are processed to determine the reconstructed migration image of each frame. The reconstructed migration images from each frame are stitched together to obtain the target video.
2. The method according to claim 1, characterized in that, Determining the first human pose corresponding to the original image and the second human pose corresponding to each frame of the sample video based on the original image and each frame of the sample video includes: The sample video is divided into frames to obtain multiple sample images; Depth estimation is performed on the original image and each of the sample images to obtain a first depth map corresponding to the original image and a second depth map corresponding to each of the sample images. Human pose estimation is performed on the first depth map and the second depth map respectively to obtain the first human pose corresponding to the original image and the second human pose corresponding to each sample image frame.
3. The method according to claim 2, characterized in that, The original image, the first human pose, and the second human pose corresponding to each sample image are processed to determine the reconstructed migration image for each frame, including: Calculate the transformation matrix based on the first human posture and each second human posture; Based on the transformation matrix, the original image is converted into a migration map for each frame; For each frame of the migration map, image reconstruction is performed based on the migration map, the original image, the second depth map, and the second human pose to obtain multiple frames of reconstructed migration images.
4. The method according to claim 3, characterized in that, For each frame of the migration map, image reconstruction is performed based on the migration map, the original image, the second depth map, and the second human pose to obtain a reconstructed migration image, including: The original image is divided into a background image and a foreground image; The background image is processed using the acquired background generation network to obtain a reconstructed background image; For each frame of the migration map, the acquired graph generation network is used to generate a pose migration image from the migration map, the second depth map, and the second human pose, thus obtaining a pose migration foreground image. The reconstructed migration image is obtained by integrating the pose migration foreground image with the reconstructed background image.
5. The method according to claim 4, characterized in that, Also includes: Image reconstruction is performed based on the original image, the first depth map, and the first human pose to obtain the reconstructed original image; The graph generation network is updated based on the similarity between the reconstructed original image and the original image.
6. The method according to claim 3, characterized in that, Image reconstruction is performed based on the original image, the first depth map, and the first human pose to obtain the reconstructed original image, including: The original image is divided into a background image and a foreground image; The background image is processed using the acquired background generation network to obtain a reconstructed background image; The obtained graph generation network is used to generate an image from the foreground image, the first depth map, and the first human pose to obtain the first foreground image; The reconstructed original image is obtained by integrating the first foreground image with the reconstructed background image.
7. An agricultural product video generation device, characterized in that, The device includes: The acquisition unit is used to acquire the sample video selected by the user and the uploaded original image; The human pose generation unit is used to determine the first human pose corresponding to the original image and the second human pose corresponding to each frame of the sample video based on the original image and each frame of the sample video. The processing unit is used to process the original image, the first human pose, and the second human pose corresponding to each frame of sample image to determine the reconstructed migration image of each frame. The video generation unit is used to stitch together the reconstructed migration images of each frame to obtain the target video.
8. The apparatus according to claim 7, characterized in that, The human posture generation unit is specifically used for: The sample video is divided into frames to obtain multiple sample images; Depth estimation is performed on the original image and each of the sample images to obtain a first depth map corresponding to the original image and a second depth map corresponding to each of the sample images. Human pose estimation is performed on the first depth map and the second depth map respectively to obtain the first human pose corresponding to the original image and the second human pose corresponding to each sample image frame.
9. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory being used to store program code and data for data generation, and the processor being used to call program instructions in the memory to execute the agricultural product video method as described in any one of claims 1-6.
10. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the storage medium is located to perform the agricultural product video method as described in any one of claims 1-6.