Research map / Point clouds / Point cloud classification
T-MAE: Temporal Masked Autoencoders for Point Cloud Representation Learning
Weijie Wei, Fatemeh Karimi Nejadasl, Theo Gevers, Martin R. Oswald
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add T-MAE pre-training to SiamWCA | SiamWCA · Waymo val (5% labeled) | L2 mAPH: 40.29 -> 49.46 | Improved |
| add T-MAE pre-training to SiamWCA | SiamWCA · Waymo val (100% labeled) | L2 mAPH: 69.13 -> 70.52 | Improved |
| add T-MAE pre-training | SiamWCA · ONCE val | mAP: 63.71 -> 67.00 | Improved |
| pre-train other methods (PointContrast, ProposalContrast) at 100% labeled data | Waymo detector · Waymo val (100% labeled) | L2 mAP: 71.30 -> 68.06 | Hurt |
| feed two consecutive frames at test time instead of duplicating the current frame (ablation) | T-MAE · Waymo val (5% labeled) | L2 mAP: 50.89 -> 51.47 | Improved |
| concatenate points of two consecutive frames in GD-MAE instead of single frame (ablation) | GD-MAE · Waymo split 0 (5%) | L2 mAP: 48.23 -> 47.35 | Hurt |
| use learned cross-attention fusion (T-MAE) instead of point concatenation (ablation) | GD-MAE · Waymo split 0 (5%) | L2 mAP: 47.35 -> 49.45 | Improved |
| use Siamese shared encoder instead of asymmetric encoder (ablation) | SiamWCA · Waymo split 0 (5%) | L2 mAP: 47.26 -> 49.45 | Improved |
| use full-gradient Siamese instead of SimSiam-style detached encoder (ablation) | SiamWCA · Waymo split 0 (5%) | L2 mAP: 45.58 -> 49.45 | Improved |
| use windowed cross-attention instead of concat windowed self-attention (ablation) | SiamWCA · Waymo split 0 (5%) | L2 mAP: 43.82 -> 49.45 | Improved |
| add windowed self-attention after cross-attention (ablation) | SiamWCA · Waymo split 0 (5%) | L2 mAP: 49.45 -> 47.01 | Mixed |
| align previous frame to current frame via ego-pose (ablation) | SiamWCA · Waymo (5% finetune) | L2 mAP: 41.42 -> 44.05 | Improved |
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.
- GD-MAE — GD-MAE: Generative Decoder for MAE Pre-Training on LiDAR Point Clouds (CVPR 2023) · better on 3 of 3 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
- RI-MAE: Rotation-Invariant Masked AutoEncoders for Self-Supervised Point Cloud Representation Learning (AAAI 2025 · similarity 0.89)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.88)
- BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · similarity 0.87)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.87)
- Multi-Scale Neighborhood Occupancy Masked Autoencoder for Self-Supervised Learning in LiDAR Point Clouds (CVPR 2025 · similarity 0.87)
- PSA-SSL: Pose and Size-aware Self-Supervised Learning on LiDAR Point Clouds (CVPR 2025 · similarity 0.87)
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.
