This website requires JavaScript.

SketchSampler: Sketch-based 3D Reconstruction via View-dependent Depth Sampling

Chenjian GaoQian YuLu ShengYi-Zhe SongDong Xu
Aug 2022
摘要
Reconstructing a 3D shape based on a single sketch image is challenging dueto the large domain gap between a sparse, irregular sketch and a regular, dense3D shape. Existing works try to employ the global feature extracted from sketchto directly predict the 3D coordinates, but they usually suffer from losingfine details that are not faithful to the input sketch. Through analyzing the3D-to-2D projection process, we notice that the density map that characterizesthe distribution of 2D point clouds (i.e., the probability of points projectedat each location of the projection plane) can be used as a proxy to facilitatethe reconstruction process. To this end, we first translate a sketch via animage translation network to a more informative 2D representation that can beused to generate a density map. Next, a 3D point cloud is reconstructed via atwo-stage probabilistic sampling process: first recovering the 2D points (i.e.,the x and y coordinates) by sampling the density map; and then predicting thedepth (i.e., the z coordinate) by sampling the depth values at the raydetermined by each 2D point. Extensive experiments are conducted, and bothquantitative and qualitative results show that our proposed approachsignificantly outperforms other baseline methods.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答