Research map / Detection and segmentation / 3D detection and driving
DeepInteraction: 3D Object Detection via Modality Interaction
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 |
|---|---|---|---|
| Maintain separate modality representations and interact instead of representational fusion | DeepInteraction · nuScenes val | mAP: 67.5 -> 68.7 | Helped |
| Use MMPI instead of DETR decoder layers (ablation) | DeepInteraction decoder · nuScenes val | mAP: 68.6 -> 69.9 | Helped |
| Add MMPI on the image representation while retaining DETR on LiDAR (ablation) | DeepInteraction decoder · nuScenes val | mAP: 68.6 -> 69.3 | Helped |
| Use alternating LiDAR and image representations instead of LiDAR throughout (ablation) | DeepInteraction decoder · nuScenes val | mAP: 69.2 -> 69.9 | Helped |
| Increase decoder depth from 1 to 5 layers (ablation) | DeepInteraction decoder · nuScenes val | mAP: 65.1 -> 69.9 | Helped |
| Increase decoder depth from 5 to 6 layers (ablation) | DeepInteraction decoder · nuScenes val | mAP: 69.9 -> 69.7 | Hurt |
| Use 200 training queries and 300 inference queries rather than 200 at both stages (ablation) | DeepInteraction decoder · nuScenes val | mAP: 69.9 -> 70.1 | Helped |
| Add MMRI to intra-modal learning (ablation) | DeepInteraction encoder · nuScenes val | mAP: 68.1 -> 69.9 | Helped |
| Add IML to cross-modal representational interaction (ablation) | DeepInteraction encoder · nuScenes val | mAP: 69.5 -> 69.9 | Helped |
| Add one representational interaction encoder layer (ablation) | DeepInteraction encoder · nuScenes val | mAP: 66.4 -> 67.7 | Helped |
| Increase encoder depth from 1 to 2 layers (ablation) | DeepInteraction encoder · nuScenes val | mAP: 67.7 -> 69.9 | Helped |
| Add multimodal interaction to Transfusion-L | VoxelNet-based detector · nuScenes val | mAP: 65.1 -> 69.9 | Helped |
Closest papers by meaning
- Unifying Voxel-based Representation with Transformer for 3D Object Detection (NeurIPS 2022 · similarity 0.88)
- CCF: Complementary Collaborative Fusion for Domain Generalized Multi-Modal 3D Object Detection (CVPR 2026 · similarity 0.87)
- Taming Cascaded Mixture-of-Experts for Modality-missing Multi-modal Salient Object Detection (AAAI 2026 · similarity 0.87)
- PiMAE: Point Cloud and Image Interactive Masked Autoencoders for 3D Object Detection (CVPR 2023 · similarity 0.87)
- DiMSOD: A Diffusion-Based Framework for Multi-Modal Salient Object Detection (AAAI 2025 · similarity 0.86)
- Interactron: Embodied Adaptive Object Detection (CVPR 2022 · 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.