基于手绘草图的服饰类别3D打印模型生成数据

通过构建一个包含大量不同外形、且均为水密性的服饰3D打印模型及其对应的手绘设计草图（2D线稿图）的大规模配对数据集，可以为深度学习模型提供训练基础，使其学习从二维线条轮廓生成功能性的三维物体。这一数据集主要适用于个性化办公用品和桌面收纳的定制服务、中小学设计与制造课程教学以及创意产品原型的快速验证。利用该数据训练出的模型，能够让普通用户甚至儿童通过随手画一张服饰的草图，就能生成一个可直接用于3D打印的实体模型，解决了即便是简单功能性物品也需要专业三维建模软件才能设计，以及普通人无法将创意草图快速实物化的问题。基于手绘草图生成特定类别（如服饰）的可3D打印模型，旨在实现创意的快速物化。具体过程包括：（1）数据收集：用户在纸上或平板电脑上绘制一张能清晰表达服饰外轮廓的2D设计草图（I_sketch）。（2）数据处理：将输入的草图图片进行预处理，然后送入一个在草图数据集上训练过的图像编码器，以提取代表设计意图的草图特征向量。特征向量通过公式 F_sketch = Encoder_sketch(I_sketch) 提取，其中 F_sketch 为草图特征向量，Encoder_sketch 为图像编码器。（3）模型构建：使用提取的草图特征向量作为输入，设计并搭建一个专注于从2D轮廓生成3D实体的几何解码模型。该模型生成服饰的隐式三维表示，并确保其具有中空结构以实现服饰功能。根据公式 SDF = Decoder_3D(F_sketch) 从草图特征中解码出三维模型，Decoder_3D 为三维形状解码器。关键的评估指标包括交并比（Intersection over Union, IoU）和倒角距离（Chamfer Distance, CD）。此方法适用于将非专业用户的2D手绘创意直接转化为功能性的3D打印实物，极大地简化了实用物品的设计流程。

Constructing a large-scale paired dataset comprising numerous water-tight 3D printable clothing models and their corresponding hand-drawn design sketches (2D line drawings) can provide a robust training foundation for deep learning models, enabling them to learn the task of generating functional three-dimensional objects from two-dimensional line contours.

This dataset is primarily applicable to customized services for personalized office supplies and desktop organizers, teaching in K-12 design and manufacturing courses, and rapid validation of creative product prototypes.

Models trained using this dataset allow ordinary users and even children to generate a directly 3D printable physical model by simply sketching a clothing design casually, addressing two core pain points: the need for professional 3D modeling software to design even simple functional items, and the inability of non-professionals to rapidly turn their creative sketches into physical objects.

Generating category-specific (e.g., clothing) 3D printable models from hand-drawn sketches aims to achieve rapid materialization of creative ideas.

The specific workflow includes:
1. Data Collection: Users draw a 2D design sketch (I_sketch) that clearly expresses the outer contour of clothing on paper or a tablet.
2. Data Processing: Preprocess the input sketch image, then feed it into an image encoder trained on sketch datasets to extract sketch feature vectors that represent design intent. The feature vector is extracted via the formula $F_{ ext{sketch}} = ext{Encoder}_{ ext{sketch}}(I_{ ext{sketch}})$, where $F_{ ext{sketch}}$ denotes the sketch feature vector and $ ext{Encoder}_{ ext{sketch}}$ is the image encoder.
3. Model Construction: Use the extracted sketch feature vectors as input to design and build a geometric decoding model dedicated to generating 3D entities from 2D contours. This model generates an implicit 3D representation of clothing and ensures it has a hollow structure to fulfill the functional requirements of the clothing. The 3D model is decoded from the sketch features via the formula $SDF = ext{Decoder}_{3D}(F_{ ext{sketch}})$, where $ ext{Decoder}_{3D}$ is the 3D shape decoder.

Key evaluation metrics include Intersection over Union (IoU) and Chamfer Distance (CD).

This method is suitable for directly converting 2D hand-drawn creative ideas from non-professional users into functional 3D printable physical objects, greatly simplifying the design workflow of practical items.

搜集汇总

背景与挑战

背景概述

该数据集包含6237条CSV格式数据，涵盖草图、特征向量和3D模型等字段，用于训练深度学习模型从手绘草图生成服饰类3D打印模型。它支持个性化定制、教学和创意验证应用，通过算法处理草图并评估生成质量，简化了非专业用户将创意转化为实物的流程。