AI Paper Atlas

Research map / Point clouds / Point cloud classification

Point Contrastive Prediction with Semantic Clustering for Self-Supervised Learning on Point Cloud Videos

Xiaoxiao Sheng, Zhiqiang Shen, Gang Xiao, Longguang Wang, Yulan Guo, Hehe Fan

ICCV 2023 · Point clouds › Point cloud classification · approach: Self-supervised point cloud pretraining · 25 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 positive neighbor sample selection to local contrastive prediction (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 91.38 -> 91.98Improved
add prototype alignment (semantic clustering pretext) (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 91.98 -> 92.26Improved
add prototype alignment and soft category alignment together (ablation)PointCPSC · Synthia 4D ablationmIoU: 70.45 -> 71.13Improved
keep only 70% of ranked negatives (drop the most similar undesired negatives) (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 90.23 -> 92.26Improved
weighted fusion of target and neighbors instead of softmax weighting (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 89.90 -> 92.26Improved
increase positive neighbors from 3 to 10 (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 92.26 -> 90.57Hurt
increase memory bank size from 256 to 1024 (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 90.91 -> 92.26Improved
increase prototypes from 10 to 20 on MSRAction-3D (ablation)PointCPSC · MSRAction-3D 16 framesaccuracy: 92.26 -> 90.91Hurt
self-supervised pre-training of PSTNet, end-to-end fine-tuned, 16 framesPSTNet · MSRAction-3Daccuracy: 89.90 -> 92.26Improved
self-supervised pre-training of P4Transformer on Synthia 4DP4Transformer · Synthia 4D (3 frames)mIoU: 83.16 -> 84.47Improved
self-supervised pre-training with 50% labels on NTU-RGBD versus full-label supervised modelsPSTNet · NTU-RGBD cross-subjectaccuracy: 90.5 (PSTNet, full labels) vs 88.0 (50% labels)No clear effect
pre-training on NTU-RGBD then transfer to NvGesturePSTNet · NvGestureaccuracy: 86.1 -> 87.3Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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