Research map / Detection and segmentation / 3D detection and driving
Equivariant Spatio-Temporal Self-Supervision for LiDAR Object Detection
Deepti Hegde, Suhas Lohit, Kuan-Chuan Peng, Michael J. Jones, Vishal M. Patel
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 spatial equivariance to temporal-only pretraining (ablation) | VoxelRCNN pretraining · KITTI 5% labels | mAP (%): 65.82 -> 66.52 | Improved |
| add temporal (scene flow) equivariance to spatial-only pretraining (ablation) | VoxelRCNN pretraining · KITTI 5% labels | mAP (%): 66.01 -> 66.52 | Improved |
| use both objectives instead of spatial-only for pedestrian (ablation) | VoxelRCNN pretraining · KITTI 5% labels, Pedestrian moderate AP | AP (%): 50.34 -> 48.55 | Hurt |
| use classification objective for rotation equivariance instead of PointInfoNCE contrastive (ablation) | SECOND pretraining · KITTI SECOND | relative mAP | Improved |
| use contrastive objective for translation and scaling equivariance instead of classification (ablation) | SECOND pretraining · KITTI SECOND | relative mAP | Improved |
| pretrain with equivariant spatio-temporal SSL with 20% labels | SECOND · KITTI 20% vs 100% no pretraining | mAP (%): 61.86 -> 65.28 | Improved |
| pretrain with the method when full labels are available | VoxelRCNN · KITTI 100% labels (Waymo pretrain) | mAP (%): 71.77 -> 70.56 | Hurt |
| pretrain with the method versus invariant STRL baseline | SECOND · KITTI 5% labels (KITTI-360 pretrain) | mAP (%): 52.94 -> 58.60 | Improved |
| pretrain with the method on KITTI Car category | SECOND / VoxelRCNN · KITTI Car AP | AP (%) | No clear effect |
| method versus ALSO generative pretraining | SECOND · KITTI 5% labels | mAP (%): 59.69 -> 58.60 | Hurt |
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.
- ALSO — ALSO: Automotive Lidar Self-Supervision by Occupancy Estimation (CVPR 2023) · better on 2 of 5 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
- Equivariant Self-Supervised Learning: Encouraging Equivariance in Representations (ICLR 2022 · similarity 0.88)
- FIND: Frequency Invariance Disentanglement for Test-Time Adaptation in LiDAR 3D Detection (NeurIPS 2026 · similarity 0.88)
- Unsupervised Learning of Local Equivariant Descriptors for Point Clouds (TPAMI 2021 · similarity 0.88)
- Towards Learning Group-Equivariant Features for Domain Adaptive 3D Detection (NeurIPS 2024 · similarity 0.86)
- Self-supervised Transformation Learning for Equivariant Representations (NeurIPS 2024 · similarity 0.86)
- SIEVE: Overcoming Topological Obstruction in Equivariant Self-Supervised Learning (NeurIPS 2026 · 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.
