Learned Volume Compression for Dynamic TSDF Sequences

Abstract

This paper introduces a novel lossy compression method for Truncated Signed Distance Function (TSDF) sequences, a common 3D representation. To date, research on compressing 3D TSDF videos has been notably scarce. Our work aims to fill this gap by proposing an approach analogous to JPEG and MPEG methodologies. Keyframes (I-Frames) are compressed using a new intraframe codec, while intermediate frames are predicted using RAFT (Recurrent All-Pairs Field Transforms) [1] and subsequently corrected through a refinement process.