AI Paper Atlas

Research map / Point clouds / Segmentation and detection

Guided Point Contrastive Learning for Semi-Supervised Point Cloud Semantic Segmentation

Li Jiang, Shaoshuai Shi, Zhuotao Tian, Xin Lai, Shu Liu, Chi-Wing Fu, Jiaya Jia

ICCV 2021 · Point clouds › Segmentation and detection · approach: Point cloud semantic segmentation · 138 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 unlabeled data with guided point contrastive lossSup-only U-Net · ScanNet V2 val (5% labeled)mIoU (%): 48.1 -> 54.8Improved
add unlabeled data with guided point contrastive lossSup-only U-Net · S3DIS Area 5 (5% labeled)mIoU (%): 45.0 -> 53.0Improved
add unlabeled data with guided point contrastive lossSup-only U-Net · SemanticKITTI val (5% labeled)mIoU (%): 34.8 -> 41.8Improved
add guided contrastive loss as auxiliary on 100% labeled dataSup-only U-Net · ScanNet V2 val (100% labeled)mIoU (%): 72.9 -> 74.0Improved
add test set to unlabeled data (transductive)Semi-sup model · SemanticKITTI test (100% labeled)mIoU (%): 67.7 -> 70.2Improved
add pseudo-label guidance to point contrastive loss (ablation)PointInfoNCE · ScanNet V2 val (20% labeled)mIoU (%): 64.4 -> 65.9Improved
add confidence guidance (ablation)Model-a · ScanNet V2 val (20% labeled)mIoU (%): 65.9 -> 66.4Improved
add category-balanced sampling (ablation)Model-b · ScanNet V2 val (20% labeled)mIoU (%): 66.4 -> 66.7Improved
add category-balanced sampling instead of random sampling (ablation)Full model · SemanticKITTI val (20% labeled)mIoU (%): 57.1 -> 58.8Improved
use partial multi-view instead of random crop (ablation)Full model · ScanNet V2 val (20% labeled)mIoU (%): 66.7 -> 63.2Hurt
add projector before contrastive loss (ablation)Full model · ScanNet V2 val (20% labeled)mIoU (%): 65.0 -> 66.7Improved
use guided contrastive learning rather than self-training (ablation)Semi-sup strategies · ScanNet V2 val (20% labeled)mIoU (%): 65.5 -> 66.7Improved

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