SVG-Icons8

可缩放矢量图形 (SVG) 在现代 2D 界面中无处不在，因为它们能够缩放到不同的分辨率。然而，尽管基于深度学习的模型成功应用于光栅化图像，但矢量图表示学习和生成的问题在很大程度上仍未得到探索。在这项工作中，我们提出了一种新颖的分层生成网络，称为 DeepSVG，用于复杂的 SVG 图标生成和插值。我们的架构有效地将高级形状与编码形状本身的低级命令分开。该网络以非自回归方式直接预测一组形状。我们通过发布一个新的大规模数据集以及一个用于 SVG 操作的开源库来介绍复杂 SVG 图标生成的任务。我们证明了我们的网络学会了准确地重建各种矢量图形，并且可以通过执行插值和其他潜在空间操作作为强大的动画工具。我们的代码位于 https://github.com/alexandre01/deepsvg。

Scalable Vector Graphics (SVG) are ubiquitous in modern 2D interfaces, as they can be scaled to various resolutions. However, despite the successful application of deep learning-based models on rasterized images, the problem of vector graphics representation learning and generation remains largely unexplored. In this work, we propose a novel hierarchical generative network named DeepSVG for complex SVG icon generation and interpolation. Our architecture effectively separates high-level shapes from the low-level commands that encode the shapes themselves. The network directly predicts a set of shapes in a non-autoregressive manner. We introduce the task of complex SVG icon generation by releasing a new large-scale dataset along with an open-source library for SVG manipulation. We demonstrate that our network learns to accurately reconstruct various vector graphics, and can serve as a powerful animation tool by performing interpolation and other latent space operations. Our code is available at https://github.com/alexandre01/deepsvg.