基于多视角简约线条图的三维简约线条图生成数据
收藏资源简介：
采集或使用特定算法获得多视角简约线条图数据，利用深度生成模型进行训练，算法可以生成三维、连续的简约线条图。这一数据适用于计算机图形学、虚拟现实和增强现实等领域，用户能够通过生成的三维线条图实现更真实的三维显示和交互，同时能补充多视角简约线条图的丢失信息，使其语义表达更加完整。在多视角下合成简约线条图，并通过算法生成三维简约线条点云。具体过程如下：（1）数据收集：使用多视角虚拟相机系统拍摄三维模型网格得到多视角线条图，并人工绘制对应的三维简约线条点云。（2）数据处理：通过图像处理算法对多视角线条图进行处理，提取F_mview = Encoder_mview(Image_mview, camera)，其中F_mview表示多视角线条图特征，Encoder_mview表示预训练图像编码器，Image_mview表示多视角线条图，camera表示相机内外参。（3）三维生成：使用F_mview生成三维线条图，公式为Synth_3D = Generator_3D(F_mview)，其中Synth_3D表示生成的三维简约线条点云，Generator_3D为深度学习生成模型。最后使用平均倒角距离评估生成质量。
Multi-view simplified line drawing data is obtained via either collection or specific algorithms, and is used to train deep generative models. These algorithms are capable of generating 3D continuous simplified line drawings. This dataset is applicable to fields such as computer graphics, virtual reality (VR), and augmented reality (AR). Users can achieve more realistic 3D display and interaction through the generated 3D line drawings, while supplementing the missing information of multi-view simplified line drawings to make their semantic expression more complete. Specifically, multi-view simplified line drawings are synthesized, and 3D simplified line point clouds are generated via algorithms. The detailed workflow is as follows: (1) Data Collection: Use a multi-view virtual camera system to capture 3D model meshes to obtain multi-view line drawings, and manually create the corresponding 3D simplified line point clouds. (2) Data Processing: Process the multi-view line drawings through image processing algorithms, and extract $F_{mview} = Encoder_{mview}(Image_{mview}, camera)$, where $F_{mview}$ represents the multi-view line drawing feature, $Encoder_{mview}$ represents the pre-trained image encoder, $Image_{mview}$ represents the multi-view line drawings, and $camera$ represents the internal and external parameters of the camera. (3) 3D Generation: Use $F_{mview}$ to generate 3D line drawings, with the formula $Synth_{3D} = Generator_{3D}(F_{mview})$, where $Synth_{3D}$ represents the generated 3D simplified line point clouds, and $Generator_{3D}$ is the deep learning generative model. Finally, the average chamfer distance is used to evaluate the generation quality.
