Semantic supervision purely through 3D shape descriptors

Helia Ghasemi, Ioana Simion, Clara I. Sánchez, Hoel Kervadec

Keywords: 3d supervision, shape descriptors, anatomical priors

Registration Requirement: Yes

Abstract:

This work presents i) an adapted memory-efficient architecture that can process a 3D scan as a single patch, even at high resolution,
ii) a demonstration that this architecture can successfully be supervised purely from 3D shape descriptors, without any voxel-wise supervision at all. This can be used as a platform to conduct future research on shape supervision.

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Submission Number: 130