Research map / Point clouds / Point cloud classification
Contrastive Predictive Autoencoders for Dynamic Point Cloud Self-Supervised Learning
Xiaoxiao Sheng, Zhiqiang Shen, Gang Xiao
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Pretrain with a longer sequence (24 rather than 12 frames) (ablation) | CPR · MSRAction3D, finetune 24 frames | accuracy (%): 91.99 -> 93.38 | Improved |
| Reconstruct a segment rather than one frame (ablation) | CPR reconstruction · MSRAction3D | accuracy (%): 91.32 -> 92.68 | Improved |
| Colorize the reconstructed segment (ablation) | CPR reconstruction · MSRAction3D | accuracy (%): 92.68 -> 93.38 | Improved |
| Increase reconstruction density from 256 to 512 points per frame (ablation) | CPR reconstruction · MSRAction3D | accuracy (%): 93.38 -> 91.99 | Hurt |
| Increase reconstruction density from 256 to 1024 points per frame (ablation) | CPR reconstruction · MSRAction3D | accuracy (%): 93.38 -> 92.36 | Hurt |
| Add temporally adjacent hard negatives (ablation) | Local contrastive prediction · MSRAction3D | accuracy (%): 91.34 -> 92.23 | Improved |
| Add global contrastive prediction to local contrastive prediction (ablation) | CPR · MSRAction3D | accuracy (%): 91.34 -> 92.55 | Improved |
| Add sequence reconstruction to contrastive prediction (ablation) | CPR · MSRAction3D | accuracy (%): 92.55 -> 93.38 | Improved |
| Use contrastive prediction and reconstruction together rather than sequence reconstruction alone (ablation) | CPR · MSRAction3D | accuracy (%): 87.10 -> 93.38 | Improved |
| Transfer CPR pretraining from NTU-RGBD 60 rather than MSRAction3D | CPR · MSRAction3D, semi-supervision with 30% MSR | accuracy (%): 86.4 -> 87.2 | Improved |
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.
- PSTNet++ — PSTNet: Point Spatio-Temporal Convolution on Point Cloud Sequences (ICLR 2021) · better on 1 of 2 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
- Masked Motion Prediction with Semantic Contrast for Point Cloud Sequence Learning (ECCV 2024 · similarity 0.92)
- Point Contrastive Prediction with Semantic Clustering for Self-Supervised Learning on Point Cloud Videos (ICCV 2023 · similarity 0.91)
- PointCMP: Contrastive Mask Prediction for Self-Supervised Learning on Point Cloud Videos (CVPR 2023 · similarity 0.89)
- Adapting Pre-trained 3D Models for Point Cloud Video Understanding via Cross-frame Spatio-temporal Perception (CVPR 2025 · similarity 0.88)
- PvNeXt: Rethinking Network Design and Temporal Motion for Point Cloud Video Recognition (ICLR 2025 · similarity 0.86)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.86)
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.
