AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Perturbed Self-Distillation: Weakly Supervised Large-Scale Point Cloud Semantic Segmentation

Yachao Zhang, Yanyun Qu, Yuan Xie, Zonghao Li, Shanshan Zheng, Cuihua Li

ICCV 2021 · Point clouds › Segmentation and detection · approach: Point cloud semantic segmentation · 172 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
add point perturbation as plain data augmentation (ablation)RandLA-Net baseline · S3DIS Area-5 (1pt)mIoU: 40.7 -> 41.1No clear effect
add point perturbation as plain data augmentation (ablation)RandLA-Net baseline · S3DIS Area-5 (1%)mIoU: 58.6 -> 59.9No clear effect
add perturbed self-distillation (ablation)RandLA-Net baseline · S3DIS Area-5 (1pt)mIoU: 40.7 -> 46.9Improved
add perturbed self-distillation (ablation)RandLA-Net baseline · S3DIS Area-5 (1%)mIoU: 58.6 -> 62.0Improved
add perturbed self-distillation (ablation)RandLA-Net baseline · S3DIS Area-5 (fully supervised)mIoU: 63.0 -> 63.9Improved
add context-aware module (ablation)PSD self-distillation · S3DIS Area-5 (1pt)mIoU: 46.9 -> 48.2Improved
add context-aware module (ablation)PSD self-distillation · S3DIS Area-5 (1%)mIoU: 62.0 -> 63.5Improved
add context-aware module (ablation)PSD self-distillation · S3DIS Area-5 (fully supervised)mIoU: 63.9 -> 65.1Improved
add full PSD (self-distillation + CAM) (ablation)RandLA-Net baseline · S3DIS Area-5 (1% labels)mIoU: 58.6 -> 63.5Improved
use PointNet++ backbone with PSD instead of Xu and Lee at lower label rate (ablation)PSD+PN2 · S3DIS Area-5 (1% labels)mIoU: 48.0 -> 51.0Improved
add PSD to RandLA-Net in weak 1pt setting vs Xu and Lee with more labels (ablation)PSD · S3DIS Area-5 (1pt)mIoU: 44.5 -> 48.2Improved
add PSD training cost (perturbed branch) (ablation)RandLA-Net · S3DIS training timetraining time (s/epoch): 216 -> 302Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 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