Empowering Vector Graphics with Consistently Arbitrary Viewing and View-dependent Visibility

A DSLR photo of a time clock, clear pointer." "A DSLR photo of a LV handbag." "A flying dragon, highly detailed." "A lamma." "A crab." "An airplane." "A DSLR photo of a football helmet." "A campling." "A DSLR photo of a stylish Air Jordan shoes." "A space shuttle." "A rotary telephone." "An expensive office chair." Figure 1. Examples of multiview vector graphics generated by our method conditioned on text prompts, the first two rows are sketch and the bottom row is iconography. Our method is capable of producing vector graphics with consistent views, well-preserved shape structures, and accurate occlusion relationships. The non-visible curves are rendered with lower opacity for visualization. Small images on the top right of the sketch results are the corresponding view rendering from the auxiliary 3DGS [17] branch as a reference for shape structures.