TWI817921B - Model modeling instruction generation method and system - Google Patents

Abstract

A method for generating model modeling instructions through a model modeling instruction generation system that stores a mapping table including multiple object tags corresponding to multiple object vectors, and an instruction set including multiple construction instructions corresponding to the corresponding object vectors. to implement, and includes: (A) using a conversion method to obtain a first vector based on an input information indicating at least one object to be constructed; (B) using a decoding model based on the first vector to obtain at least one indicating The at least one second vector of the object to be constructed; (C) for each second vector, obtain a target object label matching the second vector according to the second vector and the mapping table; (D) for each target The object tag generates a pair of target construction instructions for the target object tag based on the target object tag and the instruction set.

Description

Model modeling instruction generation method and system

The present invention relates to an instruction generation method and its system, and in particular to an instruction generation method and its system used in virtual models to construct virtual objects.

With the advancement of technologies related to the metaverse and virtual reality, current virtual scene construction technology (such as: Unreal Engine, or Unreal Engine) can present a variety of quite exquisite virtual objects, allowing users to experience imaginative and exciting experiences in the virtual world. The feeling of being deeply involved in its environment.

However, although virtual scene construction technology has been widely used in fields such as video games, entertainment and design, users still need to delve into relevant background knowledge to construct virtual scenes, or rely on the assistance of a professional design team to execute, resulting in a relatively complex situation. It is too time-consuming and costly for self-employed individuals or small businesses, and it also limits the development speed of virtual scene construction technology.

In view of this, it is necessary to find a solution that allows users who want to construct virtual scenes to intuitively and simply operate any modeling software and carry out design planning, and quickly construct their own virtual world to overcome the previous excessive time consumption and cost of problem.

Therefore, an object of the present invention is to provide a method for generating model modeling instructions that allows users to intuitively and simply generate instructions for constructing a virtual scene.

Therefore, the present invention provides a method for generating model modeling instructions, which is implemented by a model modeling instruction generation system connected to an electronic device through a communication network. The model modeling instruction generation system stores a model containing a plurality of codes corresponding to a plurality of sources. a mapping table of object tags of objects constructed by the Unreal Engine and a plurality of object vectors corresponding to the corresponding object tags, and an instruction set including a plurality of construction instructions for constructing the objects and corresponding to the object vectors, and It includes one step (A), one step (B), one step (C), and one step (D).

Step (A) is to obtain a first vector corresponding to the input information using a conversion method for converting information into an output vector based on input information from the electronic device indicating at least one object to be constructed.

Step (B) is to use a decoding model for decoding and converting a vector to be decoded into at least one decoded vector indicated by at least one object to be constructed according to the first vector to obtain at least one vector indicating the at least one decoded vector. A second vector of the object to be constructed.

Step (C) is for each second vector, according to the second vector and the mapping table, and obtain a target object label matching the second vector from the object vectors.

Step (D) is for each target object tag, generating a target construction instruction for constructing the object indicated by the target object tag in the Unreal Engine according to the target object tag and the instruction set.

Another object of the present invention is to provide a model modeling instruction generation system that allows users to intuitively and simply generate instructions for constructing virtual scenes.

Therefore, the model modeling instruction generation system of the present invention includes a communication module connected to a communication network, a storage module, and a processing module electrically connected to the communication module and the storage module.

The communication module is connected to an electronic device through the communication network.

The storage module stores a mapping table including a plurality of object tags corresponding to a plurality of objects constructed by an Unreal Engine and a plurality of object vectors corresponding to the object tags, and a mapping table including a plurality of object tags for constructing the Object and a set of instructions corresponding to the construction instructions of the corresponding object vector.

The processing module, according to the input information received by the communication module from the electronic device and indicating at least one object to be constructed, uses a conversion method for converting an information into an output vector to obtain a pair of the corresponding A first vector of input information. Based on the first vector, the processing module uses a decoding model for decoding and converting a vector to be decoded into at least one decoded vector indicated by at least one object to be constructed, to obtain At least one second vector indicating the at least one object to be constructed, for each a second vector. The processing module obtains a target object label matching the second vector from the object vectors according to the second vector and the mapping table. For each target object label, the processing module obtains a target object label according to the second vector and the mapping table. The target object tag and the instruction set generate a target construction instruction for constructing the object indicated by the target object tag in the Unreal Engine.

The effect of the present invention is to use the model modeling instruction generation system to process the received input information (for example: natural language) using the conversion method to obtain the corresponding first vector, and then use the decoding model to obtain the corresponding first vector. It should be a first vector and indicate the at least one second vector of the at least one object to be constructed that the user wants to construct. Finally, according to each second vector, through the mapping table and the instruction set, each desired object is generated. The construction object corresponds to the target construction instruction of the object in the Unreal Engine, and the Unreal Engine can construct the corresponding object according to each target construction instruction, so that the user can perform virtualization through simple and intuitive natural language expression. Scene construction design.

200:Communication network

1: Model modeling instruction generation system

11: System-side communication module

12:System side storage module

13: System side processing module

2: Electronic devices

21:Device-side communication module

22: Device input module

23:Device-side processing module

51~53: Steps

61~66: Steps

621, 622: Sub-steps

641~644: Sub-steps

Other features and effects of the present invention will be clearly presented in the embodiments with reference to the drawings, in which: Figure 1 illustrates a model modeling instruction generation system for executing one embodiment of the model modeling instruction generation method of the present invention. and an electronic device; Figure 2 is a flow chart illustrating a model training program of the embodiment of the model modeling instruction generation method of the present invention; Figure 3 is a flow chart illustrating an instruction generation program of the embodiment; Figure 4 is a flow chart, Describe the sub-steps 621 and 622 of step 62 of the instruction generating program in this embodiment; and FIG. 5 is a flow chart illustrating the sub-steps 641 to 644 of step 64 of the instruction generating program of this embodiment.

Referring to Figure 1, one embodiment of the method for generating model modeling instructions of the present invention is implemented by a model modeling instruction generating system 1. The model modeling instruction generating system 1 includes a communication network 200 and an electronic device. 2 connected system-side communication module 11, a system-side storage module 12, and a system-side processing module 13 electrically connected to the system-side communication module 11 and the system-side storage module 12.

The system-side storage module 12 stores a plurality of object tags corresponding to a plurality of objects constructed by an Unreal Engine, and a plurality of object categories related to the object tags. Each object tag is classified into one of the object categories.

The system-side storage module 12 also stores a mapping table including the object tags and a plurality of object vectors corresponding to the object tags, a mapping table including a plurality of adjustable object characteristic information and a plurality of corresponding adjustable object characteristic information. , and each is represented by a positive sign One of "+" or a negative sign "-" represents an adjustment direction of an object feature adjustment direction table, an instruction set containing a plurality of construction instructions for constructing the objects and corresponding to the object vectors, an instruction set A coding model for encoding and converting a high-dimensional vector into a low-dimensional vector, and a coding model corresponding to the encoding model for decoding and converting another low-dimensional vector into another high-dimensional vector. Decoding model, multiple training sets, and a trained supervised cornerstone model. Each training set includes at least one training object name information corresponding to a structural object to be trained and the object label corresponding to each training object name information. The mapping table also includes one of the object categories corresponding to each element in each object vector.

In this embodiment, the Unreal Engine is an Unreal Engine, but is not limited thereto; the encoding model is an End-To-End model encoder (Encoder), but is not limited thereto; and the decoding model It is the decoder of the End-To-End model, but it is not limited to this; the cornerstone model is an instruction-following language model (Instruction-Following Language Model), but it is not limited to this.

It is worth mentioning that in this embodiment, the system-side processing module 13 encodes the object tag corresponding to each object that the Unreal Engine can construct (for example: One-Hot Encoding ) and obtain the object vector corresponding to the object, thereby obtaining the mapping table, but it is not limited to this method.

It is worth mentioning that in other embodiments, the Unreal Engine can In addition to being built in other computer devices, it can also be included in the model modeling command generation system 1 architecture.

The electronic device 2 includes a device-side communication module 21 connected to the model modeling instruction generation system 1 via the communication network 200, a device-side input module 22, and an electrical connection between the device-side communication module 21 and the device. The device-side processing module 23 of the terminal input module 22.

In this embodiment, the implementation form of the model modeling instruction generation system 1 is, for example, a personal computer, a server or a cloud host, but is not limited thereto.

In this embodiment, the implementation form of each electronic device 2 is, for example, a personal computer, a smart phone or a tablet computer, but is not limited thereto.

The operation details of the model modeling instruction generation system 1 and the electronic device 2 will be explained below through the embodiment of the model modeling instruction generation method of the present invention. The embodiment of the modeling instruction generation method includes a model training program, and an instruction generation program.

Referring to Figure 2, the model training program is used to train a conversion method required by the instruction generation program, and includes steps 51 to 53.

In step 51, for each training object name information in each training set, the system-side processing module 13 uses the mapping table to obtain from the object vectors according to the object tag corresponding to the training object name information. A pair of training object vectors that should contain training object name information.

In step 52, for each training set, the system-side processing module 13 According to the at least one training object vector corresponding to the at least one training object name information in the training set, the coding model is used to obtain a third vector corresponding to the training set. Among them, the dimensions of each training object vector (for example: 50000 dimensions) are the same as each other and are larger than the third vector (for example: 300 dimensions).

In step 53, the system-side processing module 13 uses the training object name information corresponding to each training set and the third vector corresponding to the training set as a training data, and based on all the training data, the The cornerstone model undergoes fine-tune training to obtain a second-stage model included in the transformation method used in sub-step 622 of the instruction generation program. It is worth mentioning that the fine-tuned second-stage model is used to convert an object name into a vector of specific dimensions that is different from the data type output by the cornerstone model. Compared with the cornerstone model, it is more suitable for specific tasks or application scenarios; in this embodiment, based on the Adapter plug-in method and using all training data, the parameters of the cornerstone model are not fine-tuned, but only the parameters of the added Adapter plug-in are fine-tuned for training to obtain the second stage. model, but not limited to this.

Referring to Figure 3, the command generation program is used to generate commands for the Unreal Engine to create corresponding objects, and includes steps 61 to 66.

In step 61, the device processing module 23 responds to an input signal generated by the device input module 22 through the user's operation, and generates input information indicating at least one object to be constructed, through the device The communication module 21 will input the information The model is sent to the modeling instruction generation system 1. It is worth mentioning that in this embodiment, the input information includes natural language. In other embodiments, the input information may also include a voice, a picture, or a video.

In step 62, after receiving the input information through the system-side communication module 11, the system-side processing module 13 uses the conversion method for converting an information into an output vector according to the input information to obtain an The first vector corresponding to the input information.

Referring to Figure 4, step 62 also includes sub-steps 621 and 622.

In sub-step 621, the system-side processing module 13 uses a first-stage model for extracting at least one object information about at least one object to be constructed from the information to perform a step of the conversion method according to the input information. The information retrieval process obtains at least one object information to be constructed from the input information, and the at least one object information to be constructed corresponds to at least one object to be constructed respectively. Among them, each object information to be constructed includes a name information of the object to be constructed and a characteristic information of the object to be constructed. The input information may also include a browsing record of the user's search for the at least one object to be constructed.

It is worth mentioning that in this embodiment, the first-stage model is the cornerstone model, but it is not limited to this.

For example, if the input information is "the room is quite elegantly decorated and has a large desk", then the first set of object information to be constructed is "room (object name information to be constructed)" and "elegant (to construct object name information)" Object characteristic information)", the second group wants to The construction object information is "desk (object name information to be constructed)" and "wide (object characteristic information to be constructed)". In particular, the system-side processing module 13 can retrieve more characteristic information of the object to be constructed related to the at least one object to be constructed based on the browsing record.

In sub-step 622, the system-side processing module 13 uses a second-stage model for converting the at least one object information to the output vector according to the at least one object name information to be constructed to perform a step of the conversion method. The vector conversion program obtains the first vector corresponding to the information of the at least one object to be constructed.

In step 63, the system-side processing module 13 uses the decoding model for decoding a vector to be decoded and converting it into at least one decoded vector indicated by the at least one object to be constructed according to the first vector. , obtaining at least one second vector indicating the at least one object to be constructed. Among them, the dimensions of each second vector (for example: 50000 dimensions) are the same as each other and are larger than the dimension of the first vector (for example: 300 dimensions). For example, step 63 is used to obtain the second vectors corresponding to the "room" and "desk" in the above example.

In step 64, for each second vector, the system-side processing module 13 uses the characteristic information of the object to be constructed in the object to be constructed information corresponding to the object to be constructed corresponding to the second vector, the mapping table and the The object feature adjusts the direction table to adjust the second vector to obtain the adjusted second vector. In particular, each adjusted second vector is compared with the corresponding unadjusted second vector with the usage The appearance characteristics described by the author.

Referring to Figure 5, step 64 also includes sub-steps 641~644.

In sub-step 641, for each second vector, the system-side processing module 13 obtains an object vector to be inspected that matches the second vector from the object vectors according to the second vector and the mapping table, The object label corresponding to the object vector to be inspected is used as an object label to be inspected, and the object category to which the object label to be inspected belongs is used as a target object category. Among all the object vectors in the mapping table, the system-side processing module 13 obtains the one with the smallest vector space distance from the second vector as the object vector to be inspected.

In sub-step 642, for each second vector, the system-side processing module 13 obtains, from all elements in the second vector according to the target object type and the mapping table, an element with the same characteristics as the target object type. The element to be adjusted of the object category.

In sub-step 643, for each second vector, the system-side processing module 13 uses the object characteristic information to be constructed in the object information to be constructed corresponding to the object to be constructed corresponding to the second vector, and the object characteristics The adjustment direction table obtains a target adjustment object feature information that matches the object feature information to be constructed from the adjustable object feature information in the object feature adjustment direction table, and adjusts the target adjustment object feature information corresponding to direction as the target adjustment direction. It is worth mentioning that in this embodiment, the system-side processing module 13 uses word embedding technology to combine the characteristic information of the object to be constructed with the characteristics of each adjustable object. The information is collected for vector space similarity comparison to obtain characteristic information of the target adjustment object. Specifically, if the characteristic information of the object to be constructed is "elegant", and the word with the highest similarity to "elegant" in the adjustable object characteristic information is "elegant", then "elegant" will be used as the target adjustment object Characteristic information, and at the same time obtain the adjustment direction of the target corresponding to "Elegance".

In sub-step 644, for each second vector, the system-side processing module 13 uses a preset value and the target adjustment direction to adjust the element to be adjusted in the second vector to obtain the adjusted third Two vectors. It is worth mentioning that in this embodiment, the system-side processing module 13 determines whether to add or subtract the element to be adjusted by adding or subtracting the sign indicated by the target adjustment direction in the second vector. Default value to obtain the adjusted second vector. Specifically, if the target adjustment direction is a positive sign "+", then the element to be adjusted is added to the preset value; if the target adjustment direction is a negative sign "-", the element to be adjusted is subtracted. Go to the default value.

In step 65, for each adjusted second vector, the system-side processing module 13 obtains a matching adjusted third vector from the object vectors based on the adjusted second vector and the mapping table. A two-vector target object vector, and the object label corresponding to the target object vector is used as one target object label, and the object category to which the target object label belongs is used as another target object category. Among all the object vectors in the mapping table, the system-side processing module 13 obtains the one with the smallest vector space distance from the adjusted second vector as the target object vector.

It is worth mentioning that in this embodiment, the mapping table has recorded the object category corresponding to each element in each object vector. Therefore, after adjusting only the element to be adjusted, the adjusted third The other target object type corresponding to the two vectors will still be the same as the target object type corresponding to the second vector before adjustment. Specifically, if the label of the object to be inspected in sub-step 641 indicates "desk" and belongs to the target object category of "table", sub-step 642 obtains the object category in the second vector that also belongs to the "table" object category. The element to be adjusted, and after the element to be adjusted is adjusted in sub-step 644, the other target object category corresponding to the adjusted second vector in step 65 will still be the "table" object category. The difference is that the label of the target object corresponding to the second vector after adjustment is different from the label of the object to be viewed corresponding to the second vector before adjustment, and can better highlight what the user described. Appearance characteristics.

It is worth mentioning that in other embodiments, the system-side processing module 13 may not retrieve the characteristic information of the object to be constructed in sub-step 621, instead of executing step 64, and use the second vector ( without adjustment in step 64) to obtain the target object label.

In step 66, for each target object tag, the system-side processing module 13 generates a target for constructing the object indicated by the target object tag in the Unreal Engine according to the target object tag and the instruction set. Build instructions. In this way, the Unreal Engine can generate corresponding objects according to each target construction instruction to complete the Construction of virtual scenes.

To sum up, the model modeling instruction generation method of the present invention uses the system-side processing module 13 to extract the input information (for example: natural language) received by the user by using the first-stage model. The at least one to-be-constructed object information is constructed, and then, based on the at least one to-be-constructed object name included in the at least one to-be-constructed object information, the second-stage model is used to obtain the at least one to-be-constructed object name indicating the at least one to-be-constructed object name. the first vector, and then use the decoding model to obtain the at least one second vector respectively indicating the at least one object to be constructed according to the first vector, and then use the characteristic information of each object to be constructed to adjust the corresponding The second vector is used to obtain the adjusted second vector, and finally each adjusted second vector is generated corresponding to the target construction instruction of the object in the Unreal Engine, so that the Unreal Engine can generate the object. In this way, The user can generate the target construction and purchase instructions through a simple and intuitive dialogue description method, so that the Unreal Engine can generate various objects that conform to the characteristics described by the user to complete the ideal virtual scene, so the present invention can indeed be achieved. the goal of.

However, the above are only examples of the present invention. They cannot be used to limit the scope of the present invention. All simple equivalent changes and modifications made based on the patent scope of the present invention and the contents of the patent specification are still within the scope of the present invention. within the scope covered by the patent of this invention.

61~66: Steps

Claims (12)

A model modeling instruction generation method is implemented by a model modeling instruction generation system connected to an electronic device through a communication network. The model modeling instruction generation system stores a model containing a plurality of files corresponding to a plurality of files generated by an Unreal Engine. A mapping table between the object tag of the constructed object and a plurality of object vectors corresponding to the corresponding object tags, a mapping table containing multiple adjustable object characteristic information and multiple corresponding adjustable object characteristic information and each having one of the positive and negative signs. An object feature adjustment direction table representing the adjustment direction, and an instruction set including a plurality of construction instructions for constructing the objects and corresponding to the object vectors. The model modeling instruction generation method includes the following steps: (A) According to An input information from the electronic device indicating at least one object to be constructed, using a first-stage model for extracting at least one object information about at least one object to be constructed from an information, obtaining at least one of the input information One object information to be constructed, the at least one object information to be constructed respectively corresponds to at least one object to be constructed, each object information to be constructed includes one object name information to be constructed and one object characteristic information to be constructed; (B) According to the at least one object to be constructed To construct object name information, obtain a pair of first vectors corresponding to at least one object information to be constructed; (C) According to the first vector, use a vector to be decoded to decode and convert it into at least one object to be constructed The decoding model of the indicated at least one decoded vector obtains at least one second vector indicating the at least one object to be constructed; (D) for each second vector, using the second vector corresponding to the second vector The object characteristic information to be constructed in the object information to be constructed corresponding to the object to be constructed, the mapping table and the object characteristic adjustment direction table are used to adjust the second vector to obtain the adjusted second vector; (E) For each adjusted second vector, according to the adjusted second vector and the mapping table, a target object vector matching the adjusted second vector is obtained from the object vectors, and the target object is The object label corresponding to the vector serves as a target object label; and (F) for each target object label, generate an instruction for constructing the target object label in the Unreal Engine based on the target object label and the instruction set. The object's target construction command. The method for generating model modeling instructions as described in claim 1, wherein in step (B), according to the at least one object name information to be constructed, a third method for converting the at least one object information into the output vector is used. The two-stage model obtains the first vector corresponding to the information of at least one object to be constructed. As for the model modeling instruction generation method described in claim 2, the model modeling instruction generation system also stores multiple training sets. Each training set includes at least one training object name information corresponding to a construction object to be trained and each training set. The object label corresponding to the training object name information, wherein, before step (A), the following steps are also included: (G) For each training object name information in each training set, according to the training object name information corresponding to object labels, and use the mapping table to obtain a pair of training object vectors corresponding to the training object name information from the object vectors; (H) for each training set, according to the at least one training set in the training set Train the at least one training object vector corresponding to the object name information, and obtain a third vector corresponding to the training set using a coding model relative to the decoding model of step (C); (I) combine each training set with its The corresponding third vector is used as a training data, and a supervised cornerstone model is fine-tuned and trained based on all the training data to obtain the second-stage model of step (B); and in step (A), the The first stage model is the cornerstone model. The model modeling instruction generation method as described in claim 2, wherein the model modeling instruction generation system also stores a plurality of object categories related to the object tags, and each object tag is classified into one of the object categories. Alternatively, the mapping table also includes the object category corresponding to each element in each object vector. Step (D) also includes the following steps: (D-1) For each second vector, according to the second vector and the mapping table, from the object vectors, obtain an object vector to be inspected that matches the second vector, and use the object label corresponding to the object vector to be inspected as an object label to be inspected, and use the object vector to be inspected to View the object category to which the object label belongs as a target object category; (D-2) for each second vector, obtain a value from all elements in the second vector according to the target object category and the mapping table An element to be adjusted that has the same object type as the target object type; (D-3) for each second vector, the object to be constructed in the information of the object to be constructed corresponding to the object to be constructed corresponding to the second vector Feature information, and the object feature adjustment direction table, obtain a matching object feature information from the object feature adjustment direction table The target adjustment object feature information that is to be constructed is the object feature information, and the adjustment direction corresponding to the target adjustment object feature information is used as the target adjustment direction; and (D-4) for each second vector, use a default value and the target adjustment direction, and adjust the element to be adjusted in the second vector to obtain the adjusted second vector. The method for generating model modeling instructions as described in claim 1, wherein in step (C), the dimensions of each second vector are the same as each other and are larger than the dimension of the first vector. The method for generating model modeling instructions as described in claim 1, wherein in step (A), the input information also includes a browsing record about the at least one object to be constructed. A model modeling instruction generation system includes: a communication module connected to an electronic device via a communication network; a storage module storing a plurality of object tags corresponding to a plurality of objects constructed by an Unreal Engine and A mapping table of a plurality of object vectors corresponding to the object labels, an object characteristic including a plurality of adjustable object feature information and a plurality of adjustment directions corresponding to the adjustable object feature information and each represented by one of positive and negative signs. Adjust the direction table, and an instruction set including a plurality of construction instructions for constructing the objects and corresponding to the object vectors; a processing module electrically connected to the communication module and the storage module; and wherein, the processing The module uses input information received from the electronic device indicating at least one object to be constructed based on the communication module. Using a first-stage model for extracting at least one object information about at least one object to be constructed from an information, at least one object information to be constructed in the input information is obtained, and the at least one object information to be constructed respectively corresponds to at least An object to be constructed. Each object information to be constructed includes a name information of an object to be constructed and a characteristic information of an object to be constructed. The processing module obtains a pair of information corresponding to at least one object to be constructed based on the at least one object name information to be constructed. The first vector, the processing module uses a decoding model for decoding a vector to be decoded and converting it into at least one decoded vector indicated by the at least one object to be constructed, to obtain at least A second vector indicating the at least one object to be constructed, for each second vector, using the characteristic information of the object to be constructed in the object to be constructed corresponding to the object to be constructed corresponding to the second vector, and the mapping table and the object feature adjustment direction table, and adjust the second vector to obtain the adjusted second vector. For each adjusted second vector, the processing module uses the adjusted second vector and the adjusted second vector. The mapping table obtains a target object vector matching the adjusted second vector from the object vectors, and uses the object label corresponding to the target object vector as a target object label. For each target object label, The processing module generates a target construction instruction for constructing the object indicated by the target object tag in the Unreal Engine based on the target object tag and the instruction set. The model modeling instruction generation system of claim 7, wherein the processing module uses a second-stage model for converting the at least one object information into the output vector according to the at least one object name information to be constructed. type to obtain the first vector corresponding to the information of the at least one object to be constructed. The model modeling instruction generation system as described in claim 8, wherein the storage module also stores a plurality of training sets, each training set including at least one training object name information corresponding to a construction object to be trained and each The object label corresponding to the training object name information; wherein, for each training object name information in each training set, the processing module uses the mapping table to extract data from the object label corresponding to the training object name information. In the object vector, a training object vector corresponding to the training object name information is obtained. For each training set, the processing module uses a training object vector corresponding to the at least one training object name information in the training set. Relative to the encoding model of the decoding model, a pair of third vectors of the training set is obtained. The processing module uses each training set and the corresponding third vector as a training data, and based on all the training data, performs a supervision The cornerstone model of the formula is fine-tuned and trained to obtain the second-stage model, and the first-stage model is the cornerstone model. The model modeling instruction generation system as described in claim 8, wherein the storage module also stores a plurality of object categories related to the object tags, and each object tag is classified as one of the object categories. , the mapping table also includes the object category corresponding to each element in each object vector; wherein, for each second vector, the processing module derives from the object vectors according to the second vector and the mapping table , obtain an object vector to be inspected that matches the second vector, and align the object vector to be inspected The corresponding object label is used as an object label to be viewed, and the object category to which the object label to be viewed is used as a target object category. For each second vector, the processing module uses the target object category and the mapping to Table, from all elements in the second vector, obtain an element to be adjusted that has the same object type as the target object type. For each second vector, the processing module uses the desired object type corresponding to the second vector. The object characteristic information to be constructed in the object information to be constructed corresponding to the construction object, and the object characteristic adjustment direction table, obtain a feature matching the object characteristic to be constructed from the adjustable object characteristic information in the object characteristic adjustment direction table The target adjustment object characteristic information of the information is used, and the adjustment direction corresponding to the target adjustment object characteristic information is used as the target adjustment direction. For each second vector, the processing module uses a preset value and the target adjustment direction, Adjust the element to be adjusted in the second vector to obtain the adjusted second vector. The model modeling instruction generation system of claim 7, wherein the dimensions of each second vector are the same as each other and are larger than the dimension of the first vector. The model modeling instruction generation system of claim 7, wherein the input information also includes a browsing record of the at least one object to be constructed.