Research map / Detection and segmentation / 3D detection and driving
MAESTRO: Task-Relevant Optimization via Adaptive Feature Enhancement and Suppression for Multi-task 3D Perception
Changwon Kang, Jisong Kim, Hongjae Shin, Junseo Park, Jun Won Choi
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 CPG (ablation) | Baseline-MTL · nuScenes val (1/4 train subset, ablation) | mIoU (Occ3D occupancy): 33.6 -> 34.6 | Improved |
| add TSFG to detection only (ablation) | Baseline-MTL with CPG · nuScenes val (1/4 train subset, ablation) | mAP (3D detection): 31.3 -> 32.6 | Mixed |
| add TSFG to map only (ablation) | Baseline-MTL with CPG · nuScenes val (1/4 train subset, ablation) | mIoU (BEV map): 40.3 -> 44.0 | Mixed |
| add TSFG to all three tasks (ablation) | Baseline-MTL with CPG · nuScenes val (1/4 train subset, ablation) | mIoU (BEV map): 40.3 -> 44.1 | Improved |
| add TSFG to all three tasks (ablation) | Baseline-MTL with CPG · nuScenes val (1/4 train subset, ablation) | mIoU (Occ3D occupancy): 34.6 -> 35.9 | Improved |
| add SPA (ablation) | MAESTRO without SPA · nuScenes val (1/4 train subset, ablation) | mIoU (Occ3D occupancy): 35.9 -> 36.9 | Improved |
| add prototype-wise features (ablation) | CPG baseline · nuScenes val (1/4 train subset, ablation) | mIoU (BEV map): 40.3 -> 42.1 | Improved |
| add prototype-aware features (ablation) | TSFG · nuScenes val (1/4 train subset, ablation) | mIoU (BEV map): 42.1 -> 43.4 | Improved |
| add feature suppression (ablation) | TSFG · nuScenes val (1/4 train subset, ablation) | mIoU (BEV map): 43.4 -> 44.1 | Improved |
| add map prototypes to occupancy prototypes (ablation) | SPA · nuScenes val (1/4 train subset, ablation) | mIoU (Occ3D occupancy): 36.3 -> 36.9 | Improved |
| add detection prototypes to occupancy prototypes (ablation) | SPA · nuScenes val (1/4 train subset, ablation) | mIoU (Occ3D occupancy): 35.9 -> 36.3 | Improved |
| use MAESTRO instead of naive shared-backbone MTL | Baseline-MTL · nuScenes val | mAP (3D detection): 32.7 -> 36.4 | Improved |
Closest papers by meaning
- Towards Flexible 3D Perception: Object-Centric Occupancy Completion Augments 3D Object Detection (NeurIPS 2024 · similarity 0.84)
- CorrBEV: Multi-View 3D Object Detection by Correlation Learning with Multi-modal Prototypes (CVPR 2025 · similarity 0.84)
- 3DiffTection: 3D Object Detection with Geometry-Aware Diffusion Features (CVPR 2024 · similarity 0.83)
- Towards Generalizable Multi-Camera 3D Object Detection via Perspective Rendering (AAAI 2025 · similarity 0.83)
- DriveGEN: Generalized and Robust 3D Detection in Driving via Controllable Text-to-Image Diffusion Generation (CVPR 2025 · similarity 0.83)
- Unified Domain Generalization and Adaptation for Multi-View 3D Object Detection (NeurIPS 2024 · similarity 0.83)
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.
