SAL: Sign Agnostic Learning of Shapes From Raw Data

Figure 1: We introduce SAL: Sign Agnostic Learning for learning shapes directly from raw data, such as triangle soups (left in each gray pair; back-faces are in red). Right in each gray pair -the surface reconstruction by SAL of test raw scans; in gold -SAL latent space interpolation between adjacent gray shapes. Raw scans are from the D-Faust dataset [8].