AI Paper Atlas

Research map / Point clouds / Registration and matching

Shape Registration in the Time of Transformers

Giovanni Trappolini, Luca Cosmo, Luca Moschella, Riccardo Marin, Simone Melzi, Emanuele Rodolà

NeurIPS 2021 · Point clouds › Registration and matching · approach: Non-rigid point set registration · 0 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
apply latent-space refinement at test timeOur · FAUSTaverage geodesic error: 0.0513 -> 0.0369Improved
apply latent-space refinement at test timeOur · FAUST(1k)average geodesic error: 0.0419 -> 0.0263Improved
apply latent-space refinement at test timeOur · SHREC19average geodesic error: 0.0802 -> 0.0615Improved
apply latent-space refinement at test timeOur · FAUST(1k-outliers)average geodesic error: 0.1657 -> 0.1479Improved
use 8 rather than 32 latent vectors (ablation)Our · SURREAL and FAUSTrelative average matching error: SURREAL: 1.00 -> 0.93; FAUST: 1.00 -> 1.24Mixed
use 2 rather than 32 latent vectors (ablation)Our · SURREAL and FAUSTrelative average matching error: SURREAL: 1.00 -> 0.91; FAUST: 1.00 -> 1.29Mixed
use 4 rather than 8 encoder-decoder layers (ablation)Our · SURREAL and FAUSTrelative average matching error: SURREAL: 1.00 -> 1.04; FAUST: 1.00 -> 1.45Mixed
use 12 rather than 8 encoder-decoder layers (ablation)Our · SURREAL and FAUSTrelative average matching error: SURREAL: 1.00 -> 0.93; FAUST: 1.00 -> 1.22Mixed
use 32 rather than 64 latent dimensions (ablation)Our · SURREAL and FAUSTrelative average matching error: SURREAL: 1.00 -> 1.36; FAUST: 1.00 -> 0.82Mixed
use 128 rather than 64 latent dimensions (ablation)Our · SURREAL and FAUSTrelative average matching error: SURREAL: 1.00 -> 1.45; FAUST: 1.00 -> 0.94Mixed
replace classic point attention with density-aware surface attentionOur transformer registration model · FAUSTmatching errorImproved
train on only a sparse fraction of ground-truth correspondencesOur · registration trainingmatching quality: 10\sim20\%No clear effect

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