Research map / Point clouds / Segmentation and detection
PiMAE: Point Cloud and Image Interactive Masked Autoencoders for 3D Object Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use PiMAE pre-training | 3DETR · SUN RGB-D | AP50: 30.3 -> 33.2 | Helped |
| Use PiMAE pre-training | GroupFree3D · SUN RGB-D | AP25: 63.0 -> 64.6 | Helped |
| Use PiMAE pre-training | DETR · ScanNetV2 val | AP50: 39.8 -> 46.5 | Helped |
| Use PiMAE pre-training | MonoDETR · KITTI val, car easy | AP: 23.1 -> 26.6 | Helped |
| Use complement projection-aligned masking instead of random masking (ablation) | PiMAE with 3DETR · SUN RGB-D | AP25: 58.0 -> 59.0 | Helped |
| Use uniform projection-aligned masking instead of random masking (ablation) | PiMAE with 3DETR · SUN RGB-D | AP50: 32.9 -> 32.6 | Hurt |
| Add cross-modal reconstruction of image features from point-cloud features (ablation) | PiMAE with 3DETR · SUN RGB-D | AP25: 59.0 -> 59.4 | Helped |
| Reconstruct corresponding image pixels rather than image features (ablation) | PiMAE with 3DETR · SUN RGB-D | AP25: 59.0 -> 58.0 | Hurt |
| Jointly pre-train RGB and geometry branches instead of geometry-only pre-training (ablation) | PiMAE · SUN RGB-D | AP25: 58.4 -> 59.4 | Helped |
| Add a shared decoder for cross-modal interaction among mask tokens (ablation) | PiMAE with 3DETR · SUN RGB-D val | AP25: 58.0 -> 59.4 | Helped |
| Reduce the encoder depth from 3+3 to 2+2 layers (ablation) | PiMAE with 3DETR · SUN RGB-D val | AP25: 59.4 -> 57.5 | Hurt |
| Increase specific decoder depth from 2 to 3 layers (ablation) | PiMAE with 3DETR · SUN RGB-D val | AP25: 59.4 -> 58.1 | Hurt |
Closest papers by meaning
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.90)
- Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training (NeurIPS 2022 · similarity 0.89)
- GeoMAE: Masked Geometric Target Prediction for Self-Supervised Point Cloud Pre-Training (CVPR 2023 · similarity 0.88)
- ImOV3D: Learning Open Vocabulary Point Clouds 3D Object Detection from Only 2D Images (NeurIPS 2024 · similarity 0.87)
- 3D Feature Prediction for Masked-AutoEncoder-Based Point Cloud Pretraining (ICLR 2024 · similarity 0.87)
- MonoMAE: Enhancing Monocular 3D Detection through Depth-Aware Masked Autoencoders (NeurIPS 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.