Research map / Detection and segmentation / 3D detection and driving
UniM2AE: Multi-modal Masked Autoencoders with Unified 3D Representation for 3D Perception in Autonomous Driving
Jian Zou, Tianyu Huang, Guanglei Yang, Zhenhua Guo, Tao Luo, Chun-Mei Feng, Wangmeng Zuo
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 |
|---|---|---|---|
| interact camera and LiDAR features in 3D volume space instead of BEV during pre-training (ablation) | UniM2AE · nuScenes val detection | NDS: 64.3 -> 65.2 | Improved |
| add multi-modal interaction (MMIM) in unified space versus merging two uni-modal pre-trainings (ablation) | UniM2AE · nuScenes val detection | NDS: 63.1 -> 65.2 | Improved |
| pre-train with UniM2AE versus random initialization | fusion detector · nuScenes val, 20% labels (C+L) | mAP: 51.5 -> 55.9 | Improved |
| pre-train with PiMAE instead of random init (project points to image plane) | fusion detector · nuScenes val, 20% labels (C+L) | mAP: 51.5 -> 52.5 | No clear effect |
| transfer pre-trained MMIM when fine-tuning BEVFusion-SST (ablation) | BEVFusion-SST · nuScenes val BEV map segmentation | mIoU: 61.4 -> 67.8 | Improved |
| raise z-axis layers beyond two (ablation) | UniM2AE · nuScenes val detection | NDS: 58.3 -> 58.4 | No clear effect |
| use one z-layer instead of two (ablation) | UniM2AE · nuScenes val detection | NDS: 58.3 -> 56.6 | Hurt |
| raise masking ratio to 80% for both modalities (ablation) | UniM2AE · nuScenes val detection | mAP: 64.5 -> 63.3 | Hurt |
| add UniM2AE pre-trained weights without MMIM transfer to LiDAR-only TransFusion-L-SST | TransFusion-L-SST · nuScenes val 3D detection | NDS: 69.9 -> 70.4 | Improved |
| add UniM2AE pre-trained encoders without MMIM to BEVFusion-SST | BEVFusion-SST · nuScenes val BEV map segmentation | mIoU: 61.3 -> 61.4 | No clear effect |
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.
- FocalFormer3D — FocalFormer3D: Focusing on Hard Instance for 3D Object Detection (ICCV 2023) · better on 1 of 1 shared result
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
- DeepInteraction++: Multi-Modality Interaction for Autonomous Driving (TPAMI 2025 · similarity 0.90)
- BEV-MAE: Bird’s Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios (AAAI 2024 · similarity 0.89)
- Towards Compact 3D Representations via Point Feature Enhancement Masked Autoencoders (AAAI 2024 · similarity 0.89)
- M-BEV: Masked BEV Perception for Robust Autonomous Driving (AAAI 2024 · similarity 0.88)
- UniPAD: A Universal Pre-training Paradigm for Autonomous Driving (CVPR 2024 · similarity 0.88)
- Multi-Scale Neighborhood Occupancy Masked Autoencoder for Self-Supervised Learning in 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.
