3DFlex: displace mass or bend space?

3D Flexible Refinement
1

Following the terminology in a Gaussian splatting paper (3D Gaussian Splatting for Real-Time Radiance Field Rendering | ACM Transactions on Graphics text from section “2.3 Point-Based Rendering and Radiance Fields” in screenshot), does 3DFlex “displace mass” or “bend space”? This question about semantics is related to the Lagrangian and Eulerian specification of a flow field
Lagrangian and Eulerian specification of the flow field - Wikipedia

Screen Shot 2025-01-27 at 1.48.04 PM
Screen Shot 2025-01-27 at 1.48.04 PM696×330 37 KB

Reading about the convection operator, and the flow-field, I would say “displace mass”, i.e. Lagrangian. I think the key methods paragraph in the 3DFlex (Nature) paper is this one:

Screen Shot 2025-01-27 at 1.53.19 PM
Screen Shot 2025-01-27 at 1.53.19 PM900×716 106 KB

@rposert ?

2

Hi @Geoffrey! I am no where near an expert (or even novice) in dynamical systems, but if my understanding is correct, then yes, 3D Flex models movement as discrete parcels of density (voxels) moving from convected space to canonical space.