CN-121999121-A - Method for generating three-dimensional model by single image

CN121999121ACN 121999121 ACN121999121 ACN 121999121ACN-121999121-A

Abstract

The invention discloses a method for generating a three-dimensional model by a single image, which comprises the following steps of inputting a plurality of two-dimensional images comprising assembled products, carrying out manual marking operation on product components, establishing a component data set, training with graphic data in the component data set, establishing a semantic segmentation network model, converting graphic features of the graphic data into component images by the semantic segmentation network model, inputting images to be converted, identifying product types of the images to be converted, selecting a corresponding component data set, carrying out component segmentation by the semantic segmentation network model, separating the images to be converted into a plurality of components, and combining the plurality of components by geometric information and object description in a description file to form the three-dimensional product model.

Inventors

LI ZHERONG
Fang Yuankai

Assignees

大岳室内装修工程有限公司

Dates

Publication Date: 20260508
Application Date: 20250915
Priority Date: 20241106

Claims (10)

1. A method for generating a three-dimensional model from a single image, comprising the steps of: Inputting a plurality of two-dimensional images containing assembled products, and respectively carrying out manual marking operation on product components in the plurality of two-dimensional images to establish a component data set containing a plurality of graphic data; Training with the plurality of graphic data in the component dataset, establishing a semantic segmentation network model, the semantic segmentation network model converting graphic features of the plurality of graphic data into component images; inputting an image to be converted, identifying the product category of the image to be converted, and selecting the corresponding part data set through the product category; Performing component segmentation through the semantic segmentation network model, separating the image to be converted into a plurality of components, wherein the components respectively comprise description files, and And combining the plurality of parts through the geometric information and the object description in the description file to form a three-dimensional product model.
2. The method for generating a three-dimensional model from a single image as recited in claim 1, wherein said manual labeling comprises performing a manual back-removing process and a color block segmentation labeling on said plurality of two-dimensional images, said plurality of graphics data comprising an original graphic, a back-removed graphic and a color block labeling graphic.
3. The method of generating a three-dimensional model from a single image as recited in claim 2, wherein the manual labeling operation includes labeling the color patches with graphical additional semantic annotations including addition or removal of parts, part pattern combinations, part function transformations, part material transformations, and perceived size differences.
4. The method for generating a three-dimensional model from a single image as claimed in claim 1, wherein the semantic segmentation network model is an encoder and decoder architecture based on a conditional generation type countermeasure network, the plurality of graphic data are generated into the component image, and the output result is judged by a markov discriminator.
5. The method of generating a three-dimensional model from a single image as recited in claim 4, wherein the semantic segmentation network model includes a self-proliferation mechanism and a self-attention mechanism.
6. The method of generating a three-dimensional model from a single image as recited in claim 1, wherein the geometric information comprises contour detection results of the plurality of parts, the contour detection results comprising coordinate position and vector information, the object description comprising a relative positional relationship of the plurality of parts.
7. The method for generating a three-dimensional model from a single image as claimed in claim 1, further comprising the steps of: and carrying out refining operation on the three-dimensional product model, and modifying each part of the three-dimensional product model to form a three-dimensional fine model.
8. The method for generating a three-dimensional model from a single image as claimed in claim 1, further comprising the steps of: And importing the three-dimensional product model into three-dimensional drawing software to generate a three-dimensional drawing model corresponding to the assembled product.
9. The method of generating a three-dimensional model from a single image of claim 1, wherein the assembled product comprises a furniture product, a home appliance product, or an automotive product.
10. The method for generating a three-dimensional model from a single image as recited in claim 9, wherein the furniture item comprises a chair, a table, a bed, a sofa, or a cabinet.

Description

Method for generating three-dimensional model by single image Technical Field The present invention relates to a method for generating a three-dimensional model from a single image, and more particularly, to a method for generating a three-dimensional model from a single image of an assembled product, and a three-dimensional model generating method including a detailed model of a component of the assembled product. Background Conversion of two-dimensional (2D) images into three-dimensional (3D) objects or models is a classical problem in computer vision, which is a task that is theoretically difficult to achieve because many important geometrical properties are disturbed during the capturing of two-dimensional images, thus creating an indistinguishable problem. In order to solve the above-mentioned difficulty in conversion, the prior art may consider to utilize a synthesis manner of multiple images to perform conversion, find corresponding points in multiple images photographed at different angles, and establish a corresponding relationship of conversion to establish a three-dimensional object. However, the technology of synthesizing multiple images cannot accurately express the features of the original object, and obtaining multiple images in real life has certain difficulty. Taking the indoor design industry as an example, when planning a decoration space of a customer, an indoor designer often uses 3D modeling software (such as SktechUp, rhinoceros D, 3D Max, etc.) to design and draw, and provides simulation results in different designs through enlargement, reduction, movement, rotation, etc. of different objects, so that the customer actually experiences the content of the design proposal. However, when the customer and the indoor designer discuss the object to be newly added in the decoration space, only the picture found or published in the network or the object picture obtained by accident is often provided, it is difficult to require the customer to provide the images of the required object at various angles to perform image conversion and three-dimensional model establishment, and the indoor designer often needs to draw and establish the required object model according to the images to present the visual effect required by the customer in cooperation with the original 3D modeling software. These costs of time and labor spent modeling not only greatly increase the cost expenditure, but also reduce the work efficiency of the designer. In addition, in the conventional method of converting a two-dimensional image into a three-dimensional model, when the model of the object is built, the conversion can be performed only on the whole outline of the object in the image, if the object is an assembled product, the built three-dimensional object model cannot be disassembled according to the product components, and the model of the required component needs to be re-drawn or independently built, so that the convenience in use is difficult to improve. In view of this, although there is a technology for converting a part of two-dimensional image into a three-dimensional model, the existing conversion method requires multiple image synthesis, the conversion rate after conversion cannot be ensured, and the use of the method is limited especially for assembling products, which are difficult to disassemble and edit the component model. In view of the above, the present inventors have conceived and devised a method for generating a three-dimensional model from a single image, which aims at solving the drawbacks of the prior art and further improves the industrial implementation and utilization. Disclosure of Invention First, the technical problem to be solved In view of the above problems in the prior art, the present invention is to provide a method for generating a three-dimensional model from a single image, so as to solve the problems that the existing conversion method is difficult to accurately build the three-dimensional model and the product disassembly component model cannot be generated. (II) technical scheme According to one aspect of the present invention, a method for generating a three-dimensional model from a single image includes inputting a plurality of two-dimensional images including an assembled product, manually marking product components in the two-dimensional images, respectively, creating a component data set including a plurality of graphic data, training the plurality of graphic data in the component data set, creating a semantic segmentation network model, converting graphic features of the plurality of graphic data into component images, inputting the image to be converted, identifying a product category of the image to be converted, selecting a corresponding component data set through the product category, performing component segmentation through the semantic segmentation network model, separating the image to be converted into a plurality of components, the components respectively inclu