AI Paper Atlas

Research map / Point clouds / Shape and surface completion

Occupancy Flow: 4D Reconstruction by Learning Particle Dynamics

Michael Niemeyer, Lars Mescheder, Michael Oechsle, Andreas Geiger

ICCV 2019 · Point clouds › Shape and surface completion · approach: Unsupervised shape correspondence · 249 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use OFlow vector-field representation instead of ONet 4DOFlow · D-FAUST (seen individuals)IoU: 77.9 % -> 79.9 %Improved
add correspondence loss to trainingOFlow · D-FAUST point cloud completion (seen)IoU: 79.9 % -> 81.5 %Improved
add correspondence loss to trainingOFlow · D-FAUST point cloud completion (seen)correspondence l2 distance: 0.122 -> 0.094Improved
add correspondence loss to trainingOFlow · D-FAUST point cloud completion (unseen individual)IoU: 69.6 % -> 72.3 %Improved
add correspondence loss to trainingPSGN 4D · D-FAUST point cloud completion (seen)correspondence l2 distance: 3.234 -> 0.102Improved
use OFlow instead of ONet 4D on unseen individualOFlow · D-FAUST point cloud completion (unseen individual)Chamfer: 0.140 -> 0.095Improved
use OFlow instead of ONet 4D for inferenceOFlow · Representation power (D-FAUST)Inference time (s): 15.509 -> 0.716Improved
use OFlow instead of ONet 4DOFlow · Representation power (D-FAUST)IoU: 94.6 % -> 93.4 %No clear effect
use OFlow instead of ONet 4D on Warping CarsOFlow · Warping Cars point cloud completionIoU: 69.7 % -> 70.7 %Improved
use OFlow instead of ONet 4D on image sequencesOFlow · 4D reconstruction from images, D-FAUSTIoU: 44.0 % -> 56.6 %Improved
add correspondence loss to trainingOFlow · 4D reconstruction from images, Warping CarsIoU: 58.2 % -> 58.0 %No clear effect
use OFlow instead of nearest neighbor matchingOFlow · Shape matching, D-FAUSTcorrespondence l2 distance: 0.374 -> 0.167Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy