Engraving method based on artificial intelligence large model

Abstract

The invention provides a carving method based on an artificial intelligence large model, and relates to the technical field of artware carving, and the method comprises the following steps: accessing from a client, the client comprising a consumer terminal and a designer terminal; after the material of the carved product is selected in the client, the corresponding label or text is selected to be input and sent to the server; the server identifies the transmitted label or text, and then AIGC drawing is carried out; the graph drawn by the AIGC is fed back to the client, and when the result is not satisfied, the graph can be regenerated; after a satisfactory result is obtained, a designer can generate a 3D model through one key; and after the model is generated, the model is transmitted to engraving software, and engraving production can be carried out. The method has the beneficial effects that the material loss can be avoided, a design draft scheme with a wide range of audiences can be provided, the final visual design draft is greatly enhanced for the primary design, a large amount of workload is directly reduced, and thus the time of designers is liberated.

Description

Technical Field

[0001] The invention relates to the technical field of handicraft engraving, in particular to an engraving method based on an artificial intelligence large model. Background Art

[0002] The current production of sculpture design usually requires a lot of pre-sculpture design work before the sculpture begins, depending on the customer's needs and materials. After the sculpture designer completes this initial step, it is delivered to the customer for confirmation; after multiple rounds of demand changes and design adjustments such as "change the material, change the form, make it positive, have mountains, water", etc., it is confirmed to enter the sculpture stage.

[0003] The current operation method is to first model the design drawing, and then transfer it to the engraving machine software for machine engraving; in this process, the initial design draft usually takes 3 days. If the design is approved at one time, it takes 17 days from design to modeling and then to engraving before it can be delivered to the consumer. However, this operation is not only inefficient, but also the design draft and finished product are also affected by the designer's personal cognition, which leads to high customization costs; finally, due to the different expectations of consumers for the effect of the finished product, the return rate remains high, forming a vicious circle.

[0004] At present, the market is roughly divided into "designer hand-drawn line draft", "use drawing tool software for 2D design", then "use 3D software modeling", and finally use another software to generate tool path and transfer it to the engraving machine for production. JDPaint, ARTAM, RhinoGold 6 and other software are mainly used for design and corresponding engraving machine software for production.

[0005] The current design part of the engraving design and production process requires a lot of time to design when it comes into contact with consumers. Because of the engraver's personal level, the number of works, the craftsmanship level and the characteristics of the material, inspiration is needed for design. However, few designers have practical experience in materials and certain special characteristics, which has built a threshold for engravers. For this reason, the processing of special materials and special features has become the watershed between ordinary engravers and master engravers. Consumers can only wait for the design plan given by the engraver within 3 days, and decide on the final line draft after communicating the plan. This also requires consumers to understand the abstraction of the line draft, which raises the threshold for consumers, and the 3-day duration is no longer suitable for the current fast pace. A linear process from line draft to modeling to tool path and production has been used for a long time. Due to the small number of engravers and even fewer masters, and the need to schedule designs, the industry has produced slow and abstract drafts.

[0006] Therefore, the present application discloses an engraving method based on an artificial intelligence large model, which can be packaged as an application or used as the technical support core of other software, the Internet, etc. It mainly realizes the design and production parts in various links related to the production of engraving design, provides designers with design auxiliary references, provides consumers with design and references, and provides factories or manufacturers for production. After use, it can shorten the design time in all aspects, reduce the design time cost and labor cost, and provide multiple sets of solutions for consumers or designers to switch and choose, and can continuously learn, retrain, and continuously self-iterate the model to enhance the model; the generated renderings are converted into 3D models through interaction or automatic, and the tool paths are generated and given to the engraving machine for engraving production. Summary of the invention

[0007] The present invention overcomes the shortcomings of the prior art and provides a carving method based on artificial intelligence large models. The AIGC method can make carving designers more courageous when facing high-end materials. Model training from high-end materials will provide feasible design solutions. After the designer fully evaluates, it will be put into production, which also includes the avoidance of material loss. It can also provide a design draft solution with a wider audience when facing mass production designs, greatly reducing the initial design cycle. The 10-second solution can be replaced at any time, and the pattern texture is superimposed on the material. The final visual design draft is a great enhancement for the initial design. The generated design draft can directly generate 3D and tool path and go into production directly, which directly reduces a lot of workload and frees up the designer's time.

[0008] Through the ability of artificial intelligence large models to train industry design content, AI can directly generate renderings and 3D files for engraving production according to user requirements, thereby improving the diversity of design and the efficiency of the entire design and production process.

[0009] In order to solve the above technical problems, the present invention is achieved through the following technical solutions:

[0010] A carving method based on artificial intelligence large model includes the following steps:

[0011] S1. Access from a client, including a consumer end and a designer end;

[0012] S2. After selecting the material of the engraving product in the client, select the corresponding label or text to enter and send to the server;

[0013] S3, the server will recognize the transmitted label or text, and then perform AIGC drawing;

[0014] S4. The drawings drawn by AIGC are fed back to the client, and can be regenerated if the result is not satisfactory; after obtaining satisfactory results, the designer can generate a 3D model with one click;

[0015] S5. After the model is generated, transfer the model to the engraving software to start the engraving production.

[0016] Furthermore, the server adopts the AIGC method to reorganize the process, and then trains the AI ​​structure model, and then trains the vocabulary that appears in the traditional design method - classification, one by one, so that AI can fully understand these vocabulary; the vocabulary of the material of the engraving product is also trained using the above method; finally, this technology is released as an application or packaged as a technical service to other Internet and software carriers.

[0017] Furthermore, the model training includes data preparation and training process.

[0018] Furthermore, the data preparation includes data collection and data screening, data preprocessing and data labeling, wherein:

[0019] Data collection and data screening means: collecting about 150,000 high-quality image data through various means, eliminating pictures that do not meet the requirements, and finally screening out about 15,000 high-quality image data for use;

[0020] Data preprocessing refers to: preprocessing images, including but not limited to image cropping, scaling, color correction, background removal, normalization, etc., so that they can be input into the model; in addition, we also convert the images into Tensor format for model training;

[0021] Data labeling means: using the Tongyi Qianwen visual model to initially label these data to be trained, and then manually performing a second labeling data verification to build a high-quality data set.

[0022] Furthermore, the training process includes the following:

[0023] S1. Install the training script;

[0024] S2. Configure the pre-trained model, VAE model, load image data, annotation data, and regularized data;

[0025] S3. Set the hyperparameters for model training. The hyperparameter settings include: learning rate scheduling, optimizer optimizer, sampler batch size, data set training rounds, training accuracy, latent space learning rate, adding xformers cross attention mechanism, adding minimum signal-to-noise ratio, adding original noise offset and other parameters.

[0026] Furthermore, the screening principles for high-quality image data are as follows:

[0027] a. It should be able to clearly reflect the details of the jade texture;

[0028] b. On the basis of retaining the detailed features of the jade, other aspects should be as varied as possible; different angles, styles, etc. should be able to ensure the overall neatness of the picture, and the image should not have too much noise, blur, unclear angle, ghosting, etc.;

[0029] c. Ensure that the image still has a resolution of 512*512 or above after being cropped to the main body of the jade article;

[0030] d. Images need to be accurately classified according to jade types and shapes to facilitate subsequent training.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The technical solution proposed in the present invention will greatly reduce the time of the initial design plan. It can provide design plans for designers and can also be used by consumers to customize their own design plans. It supports text input for semantic recognition and will generate new renderings through AI for consumers or designers. The design can be repeated until satisfaction is achieved, so that the final version can be made on the same day.

[0033] 2. In addition to shortening the time of the entire process, modeling time is also saved, shortening the entire process from more than 10 days to 2 days.

[0034] 3. The accelerated process also improves the consumer experience, shortens the transaction time and increases the transaction conversion rate.

[0035] 4. Visible renderings can not only be used in production, but also as a basis for judgment when disputes arise.

[0036] 5. The continuously trained model will become more and more powerful, so that more personalized design needs can be met, and the needs that were originally placed will be met.

[0037] 6. After the renderings are generated, the designer can use the function provided by this patent to generate a 3D model that can be produced with one click; eliminating the entire process of 3D modeling, thereby greatly improving the designer's work efficiency;

[0038] 7. The generated 3D model can be directly imported into the self-developed algorithm to automatically generate tool paths and enter the engraving production stage. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are used to provide a further understanding of the present invention and are used together with the embodiments of the present invention to explain the present invention, but do not constitute a limitation of the present invention. In the accompanying drawings:

[0040] Figure 1 is a flow chart of the engraving method based on artificial intelligence large model of the present invention;

[0041] Figure 2 It is a flow chart of the design, processing and mechanical processing of modern handmade jade articles according to the present invention;

[0042] Figure 3 This is a structural reference diagram of the Stable Diffusion model described in the present invention. DETAILED DESCRIPTION

[0043] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0044] like Figure 1 As shown, the present invention claims protection for a carving method based on an artificial intelligence large model, comprising the following steps:

[0045] S1. Access from a client, including a consumer end and a designer end;

[0046] S2. After selecting the material of the engraving product in the client, select the corresponding label or text to enter and send to the server;

[0047] S3, the server will recognize the transmitted label or text, and then perform AIGC drawing;

[0048] S4. The drawings drawn by AIGC are fed back to the client, and can be regenerated if the result is not satisfactory; after obtaining satisfactory results, the designer can generate a 3D model with one click;

[0049] S5. After the model is generated, transfer the model to the engraving software to start the engraving production.

[0050] In the current production of engraving design, due to the different customer needs, materials, etc., a lot of pre-engraving design work is usually required before engraving begins. After the engraving designer completes this initial step, it is delivered to the customer for confirmation; after multiple rounds of demand changes and design adjustments such as "change the material, change the form, make it positive, have mountains, have water", etc., it is confirmed to enter the engraving stage, first the design drawing is modeled, and then it is transferred to the engraving machine software for machine engraving; this process is quite time-consuming.

[0051] Based on this, the inventors have proposed an engraving method based on a large artificial intelligence model. The method can be packaged as an application or used as the technical support core of other software, the Internet, etc. It mainly realizes the design and production parts of various links related to the production of engraving design, provides designers with design auxiliary references, provides consumers with design and references, and provides factories or manufacturers with production. After use, it can shorten the design time in all aspects, reduce the design time cost and labor cost, and provide multiple sets of solutions for consumers or designers to switch and choose, and can continuously learn, retrain, and continuously self-iterate the model to enhance the model; the generated renderings are converted into 3D models through interaction or automatic, and the tool paths are generated and given to the engraving machine for engraving production.

[0052] Specifically, the server uses AIGC to reorganize the process, then trains the AI ​​structure model, and then classifies the words that appear in the traditional design method. The words that appear in the traditional design method are such as "zodiac nature, front, mountain, water", etc. The words that appear in the traditional design method are trained one by one to allow AI to fully understand these words; the vocabulary of the material of the carving product is also trained using the above method; finally, this technology is released as an application or packaged as a technical service to other Internet and software carriers.

[0053] AIGC stands for Artificial Intelligence Generated Content, which usually refers to the process of automatically generating content using artificial intelligence technology.

[0054] like Figure 2 As shown, the AIGC method refers to our analysis and sorting out of the process of modern handmade jade design, processing and mechanical processing. Generative artificial intelligence is used to intervene in the design and production links. Among them, 3D scanning uses open source technology (no equipment is limited, all 3D scanning tools on the market can be used) to scan the jade material into a 3D model, including the basic size, surface color, volume and other parameters of the jade material. Then select a suitable surface screenshot (you can take a screenshot through the 3D scanning model, or you can directly take a picture of the jade with a mobile phone or scanner), and then the AI ​​image generation model redraws the picture based on the input jade rough picture and prompt words to generate a jade rendering.

[0055] There is also a reverse method, which is to input a picture (such as a child's portrait or a cartoon image) and generate the desired jade effect image.

[0056] Among them, 3D scanning uses open source technology (not limited to equipment, all 3D scanning tools on the market can be used) to scan jade materials into 3D models, including the basic size, surface color, volume and other parameters of the jade materials.

[0057] Then, the AI ​​image generation model is a jade image generation model based on the Diffusion Model training.

[0058] There are two types of AI image-generated 3D models. One is also a model based on image generation trained by Diffusion Model. The other is a model that generates 3D models from a single image. It uses a multi-view cross-domain attention mechanism to extract high-quality surfaces from multi-view two-dimensional representations and combines a geometry-aware normal fusion algorithm to build a high-quality three-dimensional model.

[0059] Our model is trained based on a fine-tuned stablediffusion base model. Figure 3 This is a reference diagram of the Stable Diffusion model structure. The training steps and methods are as follows:

[0060] Prerequisite: Generally, the stable diffusion model consists of three parts: a CLIP text encoder, an Image Auto Encoder-Decoder (or VAE), and a Unet structure model.

[0061] The CLIP model (Contrastive Language-Image Pre-Training) is a multimodal pre-trained neural network that can convert a variable-length text into a fixed-dimensional vector, which is used to embed the prompt text and then input into U-Net (the next model). This model is based on

[0062] openai / clip-vit-large-patch14 model.

[0063] U-net structure: As its network structure includes a stack of ResNet convolution matrices and Cross-Attention matrices. The simple explanation of Unet is that they learn the data distribution by gradually adding noise, and in turn generate new samples by gradually denoising. : The model we used contains about 3.5 billion parameters.

[0064] The VAE (Variational Auto Encoder) model is responsible for encoding the image from the image space (pixel spaceencode) to the latent space (latent space) and finally decoding it back.

[0065] Among them, model training includes data preparation and training process.

[0066] Data preparation includes data collection and data screening, data preprocessing and data labeling, among which:

[0067] Data collection and data screening means: collecting about 150,000 high-quality image data through various means, eliminating pictures that do not meet the requirements, and finally screening out about 15,000 high-quality image data for use;

[0068] Data preprocessing refers to: preprocessing images, including but not limited to image cropping, scaling, color correction, background removal, normalization, etc., so that they can be input into the model; in addition, we also convert the images into Tensor format for model training;

[0069] Data labeling means: using the Tongyi Qianwen visual model to initially label these data to be trained, and then manually performing a second labeling data verification to build a high-quality data set.

[0070] Furthermore, the training process includes the following:

[0071] S1. Install the training script (Kohya Trainer, open source, search [kohya] or [kohya ss github] online);

[0072] S2. Configure the pre-trained model, VAE model, load image data, annotation data, and regularized data;

[0073] S3. Set the hyperparameters for model training. The hyperparameter settings include: learning rate scheduling (usually cosine), optimizer (open source, AdamW), sampler (select Euler A), batch size, data set training rounds, training accuracy, latent space learning rate, adding xformers cross attention mechanism, adding minimum signal-to-noise ratio, adding original noise offset and other parameters.

[0074] In this embodiment, the screening principles for high-quality image data are as follows:

[0075] a. It should be able to clearly reflect the details of the jade texture;

[0076] b. On the basis of retaining the detailed features of the jade, other aspects should be as varied as possible; different angles, styles, etc. should be able to ensure the overall neatness of the picture, and the image should not have too much noise, blur, unclear angle, ghosting, etc.;

[0077] c. Ensure that the image still has a resolution of 512*512 or above after being cropped to the main body of the jade article;

[0078] d. Images need to be accurately classified according to jade types and shapes to facilitate subsequent training.

[0079] The artificial intelligence large model-based engraving method proposed in the present invention will greatly reduce the time of the initial design plan, can provide design plans for designers, and can also be used by consumers to customize their own design plans. It supports text input for semantic recognition and will generate new renderings through AI for consumers or designers. The design can be repeated until it is satisfactory, so that the final version can be made on the same day; it increases the consumer's consumption experience, shortens the transaction time and improves the transaction conversion rate; the visible renderings can not only be used for production, but also can be used as a basis for judgment when disputes arise; in addition, the continuously trained model will become more and more powerful, so that more personalized design needs can be met, and the part of the needs that were originally placed will be met; after generating the renderings, the designer can generate a producible 3D model with one click through the function provided by this patent; the entire set of 3D modeling process is eliminated, thereby greatly improving the designer's work efficiency.

[0080] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. However, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A carving method based on artificial intelligence large model, characterized in that: The steps include: S1. Access from a client, including a consumer end and a designer end; S2. After selecting the material of the engraving product in the client, select the corresponding label or text to enter and send to the server; S3, the server will recognize the transmitted label or text, and then perform AIGC drawing; S4. The drawings drawn by AIGC are fed back to the client, and can be regenerated if the result is not satisfactory; after obtaining satisfactory results, the designer can generate a 3D model with one click; S5. After the model is generated, it is transferred to the engraving software to start the engraving production.

2. The engraving method based on artificial intelligence large model according to claim 1 is characterized in that: The server uses AIGC to reorganize the process, then trains the AI ​​structure model, and then trains the vocabulary that appears in the traditional design method - classification, one by one, so that AI can fully understand these vocabulary; the vocabulary of the material of the engraving product is also trained using the above method; finally, this technology is released as an application or packaged as a technical service to other Internet and software carriers.

3. The engraving method based on artificial intelligence large model according to claim 2 is characterized in that: The model training includes data preparation and training process.

4. The engraving method based on artificial intelligence large model according to claim 3 is characterized in that: The data preparation includes data collection and data screening, data preprocessing and data labeling.

5. The engraving method based on artificial intelligence large model according to claim 4 is characterized in that: The data collection and data screening refer to: collecting about 150,000 high-quality image data through various means, eliminating pictures that do not meet the requirements, and finally screening out about 15,000 high-quality image data for use.

6. The engraving method based on artificial intelligence large model according to claim 4 is characterized in that: The data preprocessing refers to: preprocessing the image, including but not limited to image cropping, scaling, color correction, background removal, normalization, etc., so as to input it into the model; in addition, we also convert the image into Tensor format for model training.

7. The engraving method based on artificial intelligence large model according to claim 4 is characterized in that: The data labeling refers to: using the Tongyi Qianwen visual model to initially label these data to be trained, and then manually performing a secondary labeling data verification to build a high-quality data set.

8. The engraving method based on artificial intelligence large model according to claim 3 is characterized in that: The training process includes the following: S1. Install the training script; S2. Configure the pre-trained model, VAE model, load image data, annotation data, and regularized data; S3. Set the hyperparameters for model training. The hyperparameter settings include: learning rate scheduling, optimizer optimizer, sampler batch size, data set training rounds, training accuracy, latent space learning rate, adding xformers cross attention mechanism, adding minimum signal-to-noise ratio, adding original noise offset and other parameters.

9. The engraving method based on artificial intelligence large model according to claim 3 is characterized in that: The screening principles for high-quality image data are as follows: a. It should be able to clearly reflect the details of the jade texture; b. On the basis of retaining the detailed features of the jade, other aspects should be as varied as possible; different angles, styles, etc. should be able to ensure the overall neatness of the picture, and the image should not have too much noise, blur, unclear angle, ghosting, etc.; c. Ensure that the image still has a resolution of 512*512 or above after being cropped to the main body of the jade article; d. Images need to be accurately classified according to jade types and shapes to facilitate subsequent training.

Images