Abstract Generating a line-drawing from a 3D model is one long-standing topic of non-photorealistic rendering because it enables the description of the shape simply and selectively. However, determining the intensity of the drawn lines is relatively unfocussed. In this paper, we introduce a method to determine intensity as a property of points on a line, consisting of two fully convolutional neural networks (CNNs). Extracted lines by the first CNN are encoded as line property images with long-range line information as local values. Those are effective for the second CNN to determine line thickness. The second CNN determines line thickness from them. Finally, intensities at points on a line are determined by interpreting them. We show the trained CNN’s evaluation results and show line drawings whose lines vary according to the generated intensities.

中文翻译：

---

使用带有线属性编码的 CNN 绘制 3D 模型的风格化线条

摘要 从 3D 模型生成线条图是非真实感渲染的一个长期主题，因为它可以简单且有选择地描述形状。然而，确定绘制线的强度相对不集中。在本文中，我们介绍了一种将强度确定为线上点属性的方法，该方法由两个完全卷积神经网络 (CNN) 组成。第一个 CNN 提取的线被编码为线属性图像，其中远程线信息作为局部值。这些对于第二个 CNN 确定线条粗细是有效的。第二个 CNN 从它们确定线条粗细。最后，一条线上各点的强度是通过解释它们来确定的。我们展示了经过训练的 CNN 的评估结果，并展示了线条根据生成强度而变化的线条图。