基于合成简约线条图的CAD指令数据

通过图像处理算法合成简约线条图数据，利用深度生成模型进行训练，算法可以根据绘制的线条生成相应的CAD指令数据。这一数据适用于工业设计与制造领域，用户能够通过简单的二维线条绘制快速生成精准的CAD指令，从而提升设计效率和准确性，解决传统CAD建模过程中耗时长、操作复杂的问题。在移动触控设备上绘制线条图，通过算法解析并生成对应的CAD指令，从而加速三维建模过程，可用于工业设计与制造领域。具体过程如下：（1）数据收集：使用相关渲染软件在随机相机视点下渲染出三维模型的RGB投影图。（2）数据处理：使用Canny等算法得到合成简约线条图。再使用图像编码器提取合成简约线条图特征。公式为：F_lines = Encoder_lines(Image_lines)，其中，F_lines表示合成简约线条图特征，Encoder_lines表示预训练的图像编码器，Image_lines表示合成简约线条图。（3）模型构建：使用合成简约线条图特征作为自变量，CAD指令为因变量，设计并搭建深度学习模型。通过公式提取CAD指令：CAD_commands = Decoder_lines(F_lines)，其中，CAD_commands表示CAD指令，Decoder_lines为深度学习模型。然后将生成的CAD指令解析成CAD中间模型和CAD模型。最后使用平均Acc和平均IoU评估整体生成质量。

Minimal line drawing data is synthesized via image processing algorithms and trained using deep generative models. The algorithm can generate corresponding CAD instruction data based on hand-drawn line sketches. This dataset is tailored for the field of industrial design and manufacturing: users can quickly generate precise CAD instructions through simple 2D line drawings, thereby improving design efficiency and accuracy, and resolving the issues of lengthy time consumption and complex operations in traditional CAD modeling workflows. Drawing line sketches on mobile touch devices and using the algorithm to parse and generate matching CAD instructions can accelerate the 3D modeling process, which is also applicable to industrial design and manufacturing scenarios.

The specific workflow is as follows:
(1) Data Collection: Render RGB projection images of 3D models from random camera viewpoints using relevant rendering software.
(2) Data Processing: Use algorithms such as Canny to generate synthesized minimal line drawings. Then extract features from the synthesized minimal line drawings via an image encoder. The formula is: F_lines = Encoder_lines(Image_lines), where F_lines denotes the features of the synthesized minimal line drawings, Encoder_lines represents the pre-trained image encoder, and Image_lines refers to the synthesized minimal line drawings.
(3) Model Construction: Take the features of the synthesized minimal line drawings as the independent variable and CAD instructions as the dependent variable to design and build a deep learning model. The CAD instructions are extracted using the formula: CAD_commands = Decoder_lines(F_lines), where CAD_commands represents the generated CAD instructions, and Decoder_lines is the aforementioned deep learning model. Subsequently, parse the generated CAD instructions into CAD intermediate models and final CAD models. Finally, evaluate the overall generation quality using average accuracy (Acc) and average Intersection over Union (IoU).