AI Paper Atlas

Research map / Point clouds / Point cloud classification

PSMix: Robust Point Cloud Recognition through Spectral Domain Mixing

Xin Wei, Qin Yang, Hongji Zhao, Fei Gao, Mingrui Zhu, Nannan Wang, Xinbo Gao

ICML 2026 · Point clouds › Point cloud classification · approach: Local feature aggregation networks

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
mix point clouds in the spectral domain (PSMix-S) instead of spatial mixupDGCNN · ModelNet40-Cmean accuracy: 77.30 -> 78.21 vs PointCutMix-RImproved
add PSMix-H on top of spatial mixup methodsDGCNN/PointNeXt with PointMixup, RSMix, PointCutMix · ModelNet40-Cmean accuracy: RSMix on PointNeXt 63.48 -> 80.45Improved
add PSMix-H to RSMix on DGCNN density corruptionsRSMix · ModelNet40-C densityaccuracy: 68.37 -> 67.93Hurt
hierarchical coefficient mixing instead of uniform linear mix (ablation)PSMix-S · ModelNet40-C (PointNeXt)mean accuracy: 78.70 -> 79.10Improved
add rotation augmentation (ablation)PSMix-S · ModelNet40-C (PointNeXt)mean accuracy: 76.70 -> 79.10Improved
do rotation in spectral domain instead of spatial (ablation)PSMix-S · ModelNet40-C (PointNeXt)mean accuracy: 77.93 -> 78.68Improved
make rotation adversarial instead of random (ablation)PSMix-S · ModelNet40-C (PointNeXt)mean accuracy: 78.68 -> 79.10Improved
sample SHT degree L randomly from 15-70 instead of fixing it (ablation)PSMix-S · ModelNet40-C (PointNeXt)mean accuracy: best fixed 78.68 (L=60) -> 79.10Improved
increase fixed SHT degree L beyond 60 (ablation)PSMix-S · ModelNet40-C (PointNeXt)mean accuracy: 78.68 -> 78.07Hurt
set transition threshold tau to 5 instead of 3 (ablation)PSMix-S · ModelNet40-Cmean accuracy: peak 79.46 at tau=5Mixed
use patch-wise variant PSMix-P instead of PSMix-S (ablation)PointNeXt · ScanObjectNN-Cmean accuracy: 57.47 -> 59.07 (transformation 75.25 -> 72.77)Mixed
use PSMix-S instead of spatial mixup on Point-MAEPoint-MAE · ModelNet40-Ctraining time per epoch: 76.22s -> 114.62sHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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