AI Paper Atlas

Research map / Point clouds / Point cloud classification

Variance-Aware Weight Initialization for Point Convolutional Neural Networks

Pedro Hermosilla, Michael Schelling, Tobias Ritschel, Timo Ropinski

ECCV 2022 · Point clouds › Point cloud classification · approach: Efficient point sampling architectures · 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
use variance-aware init instead of BatchNorm at batch size 2PCCNN · ScanObjectNN (batch size 2)accuracy: 23.3 -> 85.7Improved
use variance-aware init instead of BatchNorm at batch size 16PCCNN · ScanObjectNN (batch size 16)accuracy: 87.9 -> 83.5Hurt
use variance-aware init instead of BatchNorm at batch size 2MCConv · ScanObjectNN (batch size 2)accuracy: 21.3 -> 86.3Improved
add variance-aware init (vs He init, no BN) (ablation)KPConv · ScanObjectNN (batch size 2)accuracy: 78.9 -> 84.8Improved
add variance-aware init (vs He init, no BN) (ablation)PCCNN · ScanObjectNN (batch size 2)accuracy: 13.5 -> 85.7Improved
use variance-aware init instead of Group NormalizationSPHConv · ScanObjectNN (batch size 2)accuracy: 79.8 -> 82.2Improved
use variance-aware init (direct) instead of BatchNorm at batch size 1PointConv · ScanNet val (batch size 1)mIoU: 41.4 -> 62.3Improved
add variance-aware init (vs standard init, no BN) (ablation)PointConv · ScanNet val (batch size 1)mIoU: 1.6 -> 62.3Improved
use transfer init instead of direct init (ablation)MCConv · ScanNet val (batch size 1)mIoU: 67.4 -> 67.9No clear effect
use variance-aware init instead of BatchNorm with batch 4 small cloudsSPHConv · ScanNet valmIoU: 59.5 -> 60.1No clear effect
use gradient accumulation (virtual batch 4) with BNKPConv · ScanNet valmIoU: 32.1Hurt
add variance-aware init (vs no init, no BN) on official architecture (ablation)PointConv · ScanNet val (official arch)mIoU: 1.7 -> 59.3Improved

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