Research map / 3D vision / 3D generation
Enhanced Spatiotemporal Consistency for Image-to-LiDAR Data Pretraining
Xiang Xu, Lingdong Kong, Hui Shuai, Wenwei Zhang, Liang Pan, Kai Chen, Ziwei Liu, Qingshan Liu
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 view consistency alignment (ablation) | SLiDR with VFM superpixels · nuScenes linear probing | mIoU: 44.65 -> 45.57 | Improved |
| add dense-to-sparse regularization (ablation) | SuperFlow++ (VC) · nuScenes linear probing | mIoU: 45.57 -> 46.17 | Improved |
| add intra-sensor temporal contrastive learning (ablation) | SuperFlow++ (VC) · nuScenes linear probing | mIoU: 45.57 -> 47.24 | Improved |
| add intra-sensor temporal contrastive learning (ablation) | SuperFlow++ (VC+D2S) · nuScenes linear probing | mIoU: 46.17 -> 47.66 | Improved |
| add cross-sensor temporal contrastive learning (ablation) | SuperFlow++ (VC+D2S+ITL) · nuScenes linear probing | mIoU: 47.66 -> 48.02 | Improved |
| add temporal voting at inference (ablation) | SuperFlow++ (VC+D2S+ITL+CTL) · nuScenes linear probing | mIoU: 48.02 -> 48.86 | Improved |
| use 2 sweeps instead of 1 for dense point cloud (ablation) | SuperFlow · nuScenes linear probing | mIoU: 47.41 -> 47.66 | Improved |
| use 7 sweeps instead of 2 (ablation) | SuperFlow · nuScenes linear probing | mIoU: 47.66 -> 46.03 | Hurt |
| increase 3D backbone from MinkUNet-18 to MinkUNet-50 (ablation) | SuperFlow · nuScenes linear probing | mIoU: 47.20 -> 54.11 | Improved |
| increase 3D backbone from MinkUNet-50 to MinkUNet-101 (ablation) | SuperFlow · nuScenes linear probing | mIoU: 54.11 -> 52.56 | Hurt |
| upgrade SuperFlow to SuperFlow++ | SuperFlow (ViT-B) · nuScenes linear probing | mIoU: 47.66 -> 48.86 | Improved |
| scale 2D distillation backbone from ViT-S to ViT-L | SuperFlow++ · nuScenes linear probing | mIoU: 48.57 -> 49.78 | Improved |
Closest papers by meaning
- 4D Contrastive Superflows are Dense 3D Representation Learners (ECCV 2024 · similarity 0.96)
- SynFlow: Scaling Up LiDAR Scene Flow Estimation with Synthetic Data (ECCV 2026 · similarity 0.92)
- ICP-Flow: LiDAR Scene Flow Estimation with ICP (CVPR 2024 · similarity 0.89)
- HiLoTs: High-Low Temporal Sensitive Representation Learning for Semi-Supervised LiDAR Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.88)
- STGC-NeRF: Spatial-Temporal Geometric Consistency for LiDAR Neural Radiance Fields in Dynamic Scenes (AAAI 2025 · similarity 0.88)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 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.
