Research map / Point clouds / Point cloud classification
SimIPU: Simple 2D Image and 3D Point Cloud Unsupervised Pre-training for Spatial-Aware Visual Representations
Zhenyu Li, Zehui Chen, Ang Li, Liangji Fang, Qinhong Jiang, Xianming Liu, Junjun Jiang, Bolei Zhou, Hang Zhao
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 |
|---|---|---|---|
| Pre-train with SimIPU on KITTI image–point-cloud pairs | MVXNet with Moca · KITTI validation | Overall AP3D, moderate (%): 68.58 -> 70.88 | Improved |
| Pre-train with SimIPU on Waymo image–point-cloud pairs | MVXNet · Waymo validation | Overall mAP, L1/L2 (%): 65.57/61.53 -> 66.92/63.01 | Improved |
| Pre-train with SimIPU on Waymo image–point-cloud pairs | ResNet-50 depth baseline · KITTI Eigen split | RMS: 3.575 -> 2.840 | Improved |
| Initialize SimIPU pre-training from Super-IN, then double fine-tune | ResNet-50 depth baseline · KITTI Eigen split | RMS: 2.712 -> 2.592 | Improved |
| Initialize SimIPU pre-training from Super-IN, then double fine-tune | FCOS3D · Nuscenes validation | AP (%): +0.65 | Improved |
| Add intra-modal point-cloud contrastive learning (ablation) | SimIPU · KITTI validation | Overall AP3D, moderate (%): 69.19 -> 70.88 | Improved |
| Add inter-modal feature interaction (ablation) | SimIPU · KITTI validation | Overall AP3D, moderate (%): 68.58 -> 70.88 | Improved |
| Replace Hungarian matching with greedy assignment (ablation) | SimIPU · KITTI validation | Overall AP3D, moderate (%): 70.88 -> 68.20 | Hurt |
| Increase the amount of Waymo pre-training data (ablation) | SimIPU · KITTI monocular depth estimation and Nuscenes monocular 3D detection | RMSE and AP | Improved |
| Double fine-tune from other unsupervised pre-training (MoCo-v2) instead of SimIPU | Single-modality downstream models · Monocular depth estimation and monocular 3D object detection | Convergence | Hurt |
Closest papers by meaning
- CLIP2: Contrastive Language-Image-Point Pretraining From Real-World Point Cloud Data (CVPR 2023 · similarity 0.89)
- UniPre3D: Unified Pre-training of 3D Point Cloud Models with Cross-Modal Gaussian Splatting (CVPR 2025 · similarity 0.88)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · similarity 0.87)
- TREND: Unsupervised 3D Representation Learning via Temporal Forecasting for LiDAR Perception (NeurIPS 2025 · similarity 0.86)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.86)
- MM-Point: Multi-View Information-Enhanced Multi-Modal Self-Supervised 3D Point Cloud Understanding (AAAI 2024 · 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.
