Automobile shape design method and device, computer device and automobile

Abstract

The application relates to the field of automobile design, and discloses an automobile shape design method and device, computer equipment and an automobile, the method comprising the following steps: obtaining an automobile image training set, the automobile image training set comprising a plurality of automobile line frame diagrams, the automobile line frame diagrams being divided into a plurality of functional areas, at least part of the automobile line frame diagrams being replacement line frame diagrams, and part of the functional areas of the replacement line frame diagrams being filled with replacement design elements; using a DCGAN+Attention algorithm to train the automobile image training set, so as to construct a first automobile sketch generator and obtain a first design sketch output by the first automobile sketch generator. The application uses replacement design elements to replace the functional areas of the automobile, greatly enriches the training set, saves the preparation time of the training set, quickly and massively generates various sketches through the DCGAN+Attention algorithm, greatly improves the design efficiency of the automobile shape, and reduces the design cost of the automobile shape.

Description

Technical Field

[0001] This invention relates to the field of automotive design, and more particularly to a method, apparatus, computer equipment, and automobile for automotive exterior design. Background Technology

[0002] Currently, automotive exterior design, especially front-end styling, relies on manual hand-drawing of sketches or the use of CAD software (Computer-Aided Design) to create automotive exterior models. However, these methods consume significant human and material resources and are relatively inefficient. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for automobile exterior design to address the aforementioned technical problems, so as to improve the efficiency of automobile exterior design and reduce design costs.

[0004] A method for designing the exterior of an automobile, comprising:

[0005] Obtain a training set of car images, the training set of car images includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements;

[0006] The DCGAN+Attention algorithm is used to train the training set of car images to construct a first car sketch generator and obtain the first design sketch output by the first car sketch generator.

[0007] In the above technical solution, after training the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and obtaining the first design sketch output by the first car sketch generator, the method further includes:

[0008] Several hand-drawn sketches are obtained, and the CycleGan algorithm is used to train the hand-drawn sketches and the first design sketch to construct a second car sketch generator, and the second design sketch output by the second car sketch generator is obtained.

[0009] In the above technical solution, the acquisition of a car image training set includes several car wireframes, each wireframe being divided into several functional regions. At least a portion of the car wireframes are replacement wireframes, and some functional regions of the replacement wireframes are filled with replacement design elements, including:

[0010] Acquire car images from several specified viewpoints;

[0011] Receive annotation instructions and divide the vehicle image into several functional areas according to the annotation instructions;

[0012] The wireframe extraction tool is used to extract the wireframe outline of the car image, which has been divided into several functional areas, to generate the car wireframe diagram.

[0013] In the above technical solution, after extracting the wireframe outline of the car image that has been divided into several functional regions using a wireframe extraction tool to generate the car wireframe diagram, the method further includes:

[0014] Obtain several replacement design elements;

[0015] According to the preset replacement rules, the regional elements within the functional area of ​​the automotive wireframe diagram are replaced with the replacement design elements to generate the replacement wireframe diagram.

[0016] In the above technical solution, the step of training the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and obtaining the first design sketch output by the first car sketch generator includes:

[0017] A random vector is obtained, and the random vector is deconvolved and self-attention processed by a DCGAN network structure containing an Attention structure to generate the first feature image.

[0018] The first feature image is linearly transformed and compressed using a 1x1 convolution to obtain the first function, the second function, and the third function.

[0019] Obtain the first output feature map of the first function, and transpose the first output feature map to obtain the first transposed feature map; obtain the second output feature map of the second function; obtain the third output feature map of the third function.

[0020] Multiply the first transposed feature map with the second output feature map to obtain the second feature map;

[0021] The second feature map is normalized using an activation function to obtain the attention image;

[0022] The attention image is multiplied by the third output feature map to obtain the third feature map;

[0023] The first feature map and the third feature map are superimposed, and a deconvolution operation is performed to generate the first design sketch.

[0024] In the above technical solution, the specified viewpoint includes any one of the following: the front view of the vehicle, the rear view of the vehicle, the left side view of the vehicle, the right side view of the vehicle, and the top view of the vehicle.

[0025] In the above technical solution, the functional area includes at least one of the following: vehicle body outline, window, left rearview mirror, right rearview mirror, left headlight, right headlight, left fog light, right fog light, upper grille, and lower grille.

[0026] A vehicle exterior design device, comprising:

[0027] Obtain a training set of car images, the training set of car images includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements;

[0028] The DCGAN+Attention algorithm is used to train the training set of car images to construct a first car sketch generator and obtain the first design sketch output by the first car sketch generator.

[0029] A computer device includes a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor implements the above-described automobile exterior design method when executing the computer-readable instructions.

[0030] A car has a specific shape, and in the process of designing the specific shape, a first design sketch or a second design sketch generated by any of the above-mentioned car shape design methods is used.

[0031] The aforementioned automotive exterior design method, apparatus, computer equipment, and automobile acquire a training set of automotive images. This training set includes several automotive wireframes, each divided into functional regions. At least a portion of the wireframes are replacement wireframes, with some functional regions filled with replacement design elements, to obtain an image set that meets model training requirements. The DCGAN+Attention algorithm is used to train the automotive image training set to construct a first automotive sketch generator, and the first design sketch output by the generator is obtained, resulting in high-quality design sketches and improving the stylist's design efficiency. This invention uses replacement design elements to replace the functional regions of the automobile, greatly enriching the training set and saving preparation time. The DCGAN+Attention algorithm rapidly generates a large number of various sketches, significantly improving the efficiency of automotive exterior design and reducing design costs. Attached Figure Description

[0032] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0033] Figure 1 This is a schematic diagram of an application environment for an automobile exterior design method according to an embodiment of the present invention;

[0034] Figure 2 This is a schematic flowchart of an automobile exterior design method according to an embodiment of the present invention;

[0035] Figure 3 This is the process of generating a replacement wireframe diagram in one embodiment of the present invention;

[0036] Figure 4a This is a design sketch generated using the DCGAN algorithm in one embodiment of the present invention;

[0037] Figure 4b This is a first design sketch generated using the DCGAN+Attention algorithm in one embodiment of the present invention;

[0038] Figure 5 This is the process of generating the second design sketch in one embodiment of the present invention;

[0039] Figure 6 This is a schematic diagram of the generator and discriminator performing data processing in one embodiment of the present invention;

[0040] Figure 7 This is a schematic diagram of a DCGAN network structure including an Attention structure performing data processing in one embodiment of the present invention;

[0041] Figure 8 This is a schematic diagram of a car exterior design device according to an embodiment of the present invention;

[0042] Figure 9 This is a schematic diagram of a computer device according to an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] The automotive exterior design method provided in this embodiment can be applied to, for example... Figure 1 In this application environment, the client communicates with the server. Clients include, but are not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0045] In one embodiment, such as Figure 2 As shown, a method for designing the exterior of an automobile is provided, which is then applied to... Figure 1 Taking the server-side as an example, the explanation includes the following steps:

[0046] S10. Obtain a training set of car images, the training set of car images includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements.

[0047] Understandably, the vehicle image training set includes several vehicle wireframe images. To ensure the quality of the first vehicle sketch generator, the number of vehicle wireframe images in the training set can be greater than 100,000. In some examples, a certain angular deviation is allowed for the frontal view of the vehicle's front end.

[0048] A replacement wireframe refers to a car wireframe in which some functional areas are replaced by alternative design elements. These alternative design elements can be any design element chosen based on actual needs, such as a tiger's eyes or nose. In one example, such as... Figure 3 As shown, the left and right headlights can be replaced with the tiger's left and right eyes, respectively, and the upper grille can be replaced with the tiger's nose.

[0049] If the number of car wireframes without replacement design elements is 2000, and the number of replacement design elements in a certain functional area is 10, then 10*2000 car wireframes can be generated. As the number of replacement design elements and the number of functional areas selected increase, the number of replacement wireframes will grow exponentially, which can meet the image quantity requirements of the car image training set.

[0050] S20. The DCGAN+Attention algorithm is used to train the car image training set to construct a first car sketch generator, and the first design sketch output by the first car sketch generator is obtained.

[0051] Understandably, DCGAN (Deep Convolutional Generative Adversarial Networks) is a generative adversarial network based on deep convolutional neural networks. The DCGAN+Attention algorithm refers to introducing a self-attention structure (Attention) onto DCGAN, replacing the traditional convolutional feature maps with self-attention feature maps. Furthermore, existing DCGAN algorithms can be used to train on a car image training set. When the training effect meets the requirements, that is, when the generator's output and the discriminator's output reach a Nash equilibrium, a first car sketch generator can be obtained. A large number of first design sketches can be generated using this first car sketch generator. These first design sketches are design sketches that incorporate features from various alternative design elements. Stylists can further create based on these first design sketches, improving design efficiency and quality.

[0052] Here, the DCGAN+Attention algorithm uses fewer parameters to obtain the global geometric features of the image, focusing more on the image's contours. As shown in Figure 4, Figure 4a This is a design sketch output after training based on the DCGAN algorithm. Figure 4b This is a design sketch output after training based on the DCGAN algorithm. Figure 4a The quality is poor, characterized by uneven lines and some missing parts. Figure 4b The quality is relatively good, characterized by smooth lines and a basically complete outer contour.

[0053] In steps S10-S20, a car image training set is obtained. This training set includes several car wireframes, each divided into functional regions. At least a portion of the car wireframes are replacement wireframes, with some functional regions filled with replacement design elements, to obtain an image set that meets the model training requirements. The DCGAN+Attention algorithm is used to train the car image training set to construct a first car sketch generator. The first design sketch output by the first car sketch generator is then obtained to acquire a design sketch with higher image quality, thereby improving the stylist's design efficiency.

[0054] Optionally, after step S20, that is, after training the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and obtaining the first design sketch output by the first car sketch generator, the method further includes:

[0055] S30. Obtain several hand-drawn sketches, use the CycleGan algorithm to train the hand-drawn sketches and the first design sketch to construct a second car sketch generator, and obtain the second design sketch output by the second car sketch generator.

[0056] Understandably, hand-drawn sketches can be a stylist's personal drafts. These sketches can reflect the stylist's personal style. In some examples, the number of hand-drawn sketches can be around 1000. The CycleGAN (Cycle Generative Adversarial Networks) algorithm is a variant of the GAN algorithm. The CycleGAN algorithm uses a bidirectional discriminator to perform style transfer between a set of hand-drawn sketches (including several hand-drawn sketches) and a set of first design sketches (including several first design sketches). After training, a second car sketch generator can be obtained. The second car sketch generator can output second design sketches with the stylist's personal style. Figure 5 As shown, column a is the alternative wireframe diagram of the car, column b is the first design sketch output by the first car sketch generator, column c is the hand-drawn sketch, and column d is the second design sketch.

[0057] Optionally, step S10, namely obtaining the car image training set, the car image training set includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements, including:

[0058] S101. Obtain car images from several specified viewpoints.

[0059] Understandably, car images belonging to a specified viewpoint can be filtered from publicly available datasets. The specified viewpoint can be set according to actual needs.

[0060] In one example, the specified viewpoint includes any one of the following: the front view of the vehicle, the rear view of the vehicle, the left side view of the vehicle, the right side view of the vehicle, and the top view of the vehicle.

[0061] S102. Receive annotation instructions and divide the vehicle image into several functional areas according to the annotation instructions.

[0062] Understandably, annotation instructions can be manually entered or generated using machine recognition. Based on these annotation instructions, several functional regions can be divided on a vehicle image, with each functional region corresponding to a vehicle component or body outline.

[0063] In one example, the functional area includes at least one of the following: vehicle body outline, window, left rearview mirror, right rearview mirror, left headlight, right headlight, left fog light, right fog light, upper grille, and lower grille.

[0064] Understandably, when the specified viewpoint is the front of the vehicle, the functional areas include the vehicle's exterior outline, windows, left rearview mirror, right rearview mirror, left headlight, right headlight, left fog light, right fog light, upper grille, and lower grille.

[0065] S103. The car image, which has been divided into several functional areas, is extracted using a wireframe extraction tool to generate the car wireframe diagram.

[0066] Understandably, wireframe extraction tools can extract the wireframe outline of a car from a car image according to preset image processing rules, generating a car wireframe diagram. The car wireframe diagram can significantly reduce the amount of data processing required for training the DCGAN+Attention algorithm, thus improving training efficiency.

[0067] like Figure 3 As shown, Figure 3 (a1) is the original car image, which has been processed using a wireframe extraction tool to first transform the car image into... Figure 3 (b1) The middle image on the left, then generated Figure 3 (b1) The car wireframe diagram on the right.

[0068] Optionally, after step S103, that is, after extracting the wireframe outline of the car image that has been divided into several functional regions using a wireframe extraction tool to generate the car wireframe diagram, the method further includes:

[0069] S104. Obtain several replacement design elements;

[0070] S105. Replace the regional elements within the functional area of ​​the automotive wireframe diagram with the replacement design elements according to the preset replacement rules, and generate the replacement wireframe diagram.

[0071] Understandably, replacement design elements can be elements chosen by the stylist based on actual design needs. In one example, the replacement design element could be derived from a real tiger photograph. Figure 3 (a2) is the original tiger head image. After processing with a wireframe extraction tool, the tiger head image is first transformed into... Figure 3 (b2) The middle image on the left, then regenerated Figure 3 (b2) Wireframe of the tiger's eyes and nose on the right.

[0072] Preset replacement rules can be set according to actual needs. Preset replacement rules define which functional area is replaced by a certain type or type of replacement design element. For example, in... Figure 3 In the example, Figure 3 (b1) The right left headlight was... Figure 3 (b2) The tiger's left eye was replaced on the right. Figure 3 (b1) The right headlight of the car was... Figure 3 (b2) The tiger's right eye was replaced. Figure 3 (b1) The upper grille on the right is Figure 3 (b2) The tiger's nose on the right was replaced, ultimately forming Figure 3 The replacement wireframe diagram shown in (c) is as follows.

[0073] Understandably, step S20, namely, training the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and obtaining the first design sketch output by the first car sketch generator, includes:

[0074] S201. Obtain a random vector, and perform deconvolution and self-attention processing on the random vector through a DCGAN network structure containing an Attention structure to generate the first feature image.

[0075] S202. Perform linear transformation and channel compression on the first feature image using 1x1 convolution to obtain the first function, the second function, and the third function;

[0076] S203. Obtain the first output feature map of the first function output, and transpose the first output feature map to obtain the first transposed feature map; obtain the second output feature map of the second function output; obtain the third output feature map of the third function output;

[0077] S204. Multiply the first transposed feature map with the second output feature map to obtain the second feature map;

[0078] S205. Normalize the second feature map using an activation function to obtain an attention image;

[0079] S206. Multiply the attention image with the third output feature map to obtain the third feature map;

[0080] S207. Superimpose the first feature image and the third feature image, and perform a deconvolution operation to generate the first design sketch.

[0081] Understandably, steps S201-S207 mainly involve the process of generating a first design sketch using a first car sketch generator. Here, the first car sketch generator is a pre-trained generator. The training of the first car sketch generator and the training of the discriminator can refer to the processing methods of existing DCGAN algorithms, and will not be elaborated here.

[0082] For reference Figure 6 In step S201, a 100-dimensional random vector is obtained. The trained DCGAN network structure containing an attention structure is used to perform three deconvolution operations on the random vector, followed by a self-attention operation to generate the first feature image. The first feature image can be represented by X.

[0083] For reference Figure 7 In step S202, a 1x1 convolution can be used to perform linear transformation and channel compression on the first feature image X to obtain the first function f(x), the second function g(x), and the third function h(x).

[0084] In step S203, the first output feature map F(x) of f(x) is obtained, and the first output feature map is transposed to obtain the first transposed image F(x). T ; Obtain the second output feature map G(x) of g(x); Obtain the third output feature map H(x) of h(x).

[0085] In step S204, the first transposed image F(x) is made... T Multiplying this by the second output feature map G(x) yields the second feature map F(x). T ×G(x).

[0086] In step S205, the activation function can be the softmax function. Normalizing the second feature image using the softmax function yields the attention map (denoted by β). That is, β = soft max(F(x)). T ×G(x)).

[0087] In step S206, the attention image β is multiplied pixel by pixel with the third output feature map H(x) to obtain the third feature map o. The third feature map o is the feature map of adaptive attention, which can be expressed by the formula: o = β × H(x).

[0088] In step S207, the first feature image X and the third feature image o are superimposed, and a deconvolution operation is performed (see...). Figure 6 This process generates a first design sketch Y. In some examples, when generating the first design sketch Y, the third feature image o needs to be multiplied by a learnable factor γ. Specifically, Y = γ × o + X. The initial value of the learnable factor γ is 0. In one example, the size of the first design sketch Y can be 64px * 64px.

[0089] Before training is complete, the value of the learnable factor γ will continuously change as training progresses. For example... Figure 6As shown, the discriminator's operation is the inverse operation of the generator's generation of the sketch Y. The value generated by the final convolution operation can be obtained, and this value is used to determine whether the current sketch Y is true or false. If the sketch Y is an image generated by the generator, then the sketch Y is false; if the sketch Y is an image from the training set, then the sketch Y is true.

[0090] Since a large number of random vectors can be generated in step S201, a large number of first design sketches can be generated through the processing in steps S201-S207. The stylist can select a suitable sketch from these first design sketches and further optimize it to design a car shape that meets the requirements.

[0091] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0092] In one embodiment, an automobile exterior design apparatus is provided, which corresponds one-to-one with the automobile exterior design method described in the above embodiments. For example... Figure 8 As shown, the car exterior design device includes a training set acquisition module 10 and a first design sketch acquisition module 20. Detailed descriptions of each functional module are as follows:

[0093] The training set acquisition module 10 is used to acquire a car image training set, which includes several car wireframes. The car wireframes are divided into several functional areas. At least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements.

[0094] The first design sketch acquisition module 20 is used to train the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and acquire the first design sketch output by the first car sketch generator.

[0095] Optional, the vehicle exterior design device also includes:

[0096] A second design sketch acquisition module is used to acquire several hand-drawn sketches, and to train the hand-drawn sketches and the first design sketches using the CycleGan algorithm to construct a second car sketch generator, and to acquire the second design sketches output by the second car sketch generator.

[0097] Optionally, the training set acquisition module 10 includes:

[0098] The vehicle image acquisition unit is used to acquire vehicle images from several specified viewpoints;

[0099] A functional area division unit is used to receive annotation instructions and divide the vehicle image into several functional areas according to the annotation instructions.

[0100] The vehicle wireframe generation unit is used to extract the wireframe outline of the vehicle image, which has been divided into several functional areas, using a wireframe extraction tool to generate the vehicle wireframe.

[0101] Optionally, the training set acquisition module 10 also includes:

[0102] Get replacement element unit, used to obtain several replacement design elements;

[0103] The replacement wireframe unit is used to replace the regional elements within a functional area of ​​the automotive wireframe with the replacement design elements according to a preset replacement rule, thereby generating the replacement wireframe.

[0104] Optionally, the module 20 for obtaining the first design sketch includes:

[0105] The first feature image unit is used to obtain a random vector and convolve the car wireframe diagram according to the random vector to obtain the first feature image.

[0106] A convolutional unit is used to perform linear transformation and channel compression on the first feature image using 1x1 convolution to obtain a first function, a second function, and a third function.

[0107] The output feature map unit is used to obtain the first output feature map of the first function and transpose the first output feature map to obtain the first transposed image; obtain the second output feature map of the second function; and obtain the third output feature map of the third function.

[0108] The second feature image unit is used to multiply the first transposed image with the second output feature map to obtain the second feature image;

[0109] An attention image unit is used to normalize the second feature image using an activation function to obtain an attention image;

[0110] The third feature image unit is used to multiply the attention image with the third output feature map to obtain the third feature image;

[0111] The sketch unit to be evaluated is used to overlay the first feature image and the third feature image to obtain the sketch to be evaluated.

[0112] The discrimination unit is used to determine the authenticity of the sketch to be evaluated by a discriminator;

[0113] A first design sketch unit is configured to set the sketch to be evaluated as the first design sketch if the sketch to be evaluated is true.

[0114] Optionally, the functional area includes at least one of the following: vehicle exterior outline, windows, left rearview mirror, right rearview mirror, left headlight, right headlight, left fog light, right fog light, upper grille, and lower grille.

[0115] Specific limitations regarding the automotive exterior design device can be found in the limitations of the automotive exterior design method described above, and will not be repeated here. Each module in the aforementioned automotive exterior design device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0116] A car has a specific shape, and in the process of designing the specific shape, a first design sketch or a second design sketch generated by any of the above-mentioned car shape design methods is used.

[0117] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a readable storage medium and internal memory. The readable storage medium stores an operating system, computer-readable instructions, and a database. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the readable storage medium. The database stores data related to the automotive exterior design method. The network interface communicates with external terminals via a network connection. When the computer-readable instructions are executed by the processor, they implement an automotive exterior design method. The readable storage medium provided in this embodiment includes both non-volatile and volatile readable storage media.

[0118] In one embodiment, a computer device is provided, including a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, wherein the processor performs the following steps when executing the computer-readable instructions:

[0119] Obtain a training set of car images, the training set of car images includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements;

[0120] The DCGAN+Attention algorithm is used to train the training set of car images to construct a first car sketch generator and obtain the first design sketch output by the first car sketch generator.

[0121] In one embodiment, one or more computer-readable storage media storing computer-readable instructions are provided. The readable storage media provided in this embodiment include non-volatile readable storage media and volatile readable storage media. The readable storage media stores computer-readable instructions, which, when executed by one or more processors, perform the following steps:

[0122] Obtain a training set of car images, the training set of car images includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements;

[0123] The DCGAN+Attention algorithm is used to train the training set of car images to construct a first car sketch generator and obtain the first design sketch output by the first car sketch generator.

[0124] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for designing the exterior of an automobile, characterized in that, include: Obtain a training set of car images, the training set of car images includes several car wireframes, the car wireframes are divided into several functional areas, at least a portion of the car wireframes are replacement wireframes, and some functional areas of the replacement wireframes are filled with replacement design elements; The DCGAN+Attention algorithm is used to train the car image training set to construct a first car sketch generator and obtain the first design sketch output by the first car sketch generator. The process of acquiring a training set of car images includes several car wireframes, each wireframe being divided into several functional regions. At least a portion of the car wireframes are replacement wireframes, and some functional regions of the replacement wireframes are filled with replacement design elements, including: Acquire car images from several specified viewpoints; Receive annotation instructions and divide the vehicle image into several functional areas according to the annotation instructions; The car image, which has been divided into several functional regions, is extracted using a wireframe extraction tool to generate the car wireframe diagram. After extracting the wireframe outline of the car image, which has been divided into several functional regions, using a wireframe extraction tool to generate the car wireframe diagram, the method further includes: Obtain several replacement design elements; According to the preset replacement rules, the regional elements within the functional area of ​​the automotive wireframe diagram are replaced with the replacement design elements to generate the replacement wireframe diagram.

2. The automobile exterior design method as described in claim 1, characterized in that, After training the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and obtaining the first design sketch output by the first car sketch generator, the method further includes: Several hand-drawn sketches are obtained, and the CycleGan algorithm is used to train the hand-drawn sketches and the first design sketch to construct a second car sketch generator, and the second design sketch output by the second car sketch generator is obtained.

3. The automobile exterior design method as described in claim 1 or 2, characterized in that, The step of training the car image training set using the DCGAN+Attention algorithm to construct a first car sketch generator and obtaining the first design sketch output by the first car sketch generator includes: A random vector is obtained, and the random vector is deconvolved and self-attention processed by a DCGAN network structure containing an Attention structure to generate the first feature image. The first feature image is linearly transformed and compressed using a 1x1 convolution to obtain the first function, the second function, and the third function. Obtain the first output feature map of the first function, and transpose the first output feature map to obtain the first transposed feature map; obtain the second output feature map of the second function; obtain the third output feature map of the third function. Multiply the first transposed feature map with the second output feature map to obtain the second feature map; The second feature map is normalized using an activation function to obtain the attention image; The attention image is multiplied by the third output feature map to obtain the third feature map; The first feature map and the third feature map are superimposed, and a deconvolution operation is performed to generate the first design sketch.

4. The automobile exterior design method as described in claim 1 or 2, characterized in that, The specified viewpoint includes any one of the following: the front view of the vehicle, the rear view of the vehicle, the left side view of the vehicle, the right side view of the vehicle, and the top view of the vehicle.

5. The automobile exterior design method as described in claim 1 or 2, characterized in that, The functional area includes at least one of the following: vehicle exterior outline, windows, left rearview mirror, right rearview mirror, left headlight, right headlight, left fog light, right fog light, upper grille, and lower grille.

6. A vehicle exterior design device, characterized in that, include: The training set acquisition module is used to acquire a training set of car images. The training set of car images includes several car wireframes. The car wireframes are divided into several functional areas. At least a portion of the car wireframes are replacement wireframes. Some functional areas of the replacement wireframes are filled with replacement design elements. The module for obtaining the first design sketch is used to train the training set of the car images using the DCGAN+Attention algorithm to construct a first car sketch generator and obtain the first design sketch output by the first car sketch generator. The module for obtaining the training set includes: Acquire car images from several specified viewpoints; Receive annotation instructions and divide the vehicle image into several functional areas according to the annotation instructions; The car image, which has been divided into several functional regions, is extracted using a wireframe extraction tool to generate the car wireframe diagram. After extracting the wireframe outline of the car image, which has been divided into several functional regions, using a wireframe extraction tool to generate the car wireframe diagram, the method further includes: Obtain several replacement design elements; According to the preset replacement rules, the regional elements within the functional area of ​​the automotive wireframe diagram are replaced with the replacement design elements to generate the replacement wireframe diagram.

7. A computer device comprising a memory, a processor, and computer-readable instructions stored in the memory and executable on the processor, characterized in that, When the processor executes the computer-readable instructions, it implements the automotive exterior design method as described in any one of claims 1 to 5.

8. A car, characterized in that, The vehicle has an exterior shape, and in the process of designing the exterior shape, a first design sketch generated by the vehicle exterior design method as described in any one of claims 1 to 5 is used, or a second design sketch generated by the vehicle exterior design method as described in any one of claims 2 to 5 is used.

Images