Research map / Detection and segmentation / 3D detection and driving
RecurrentBEV: A Long-term Temporal Fusion Framework for Multi-view 3D Detection
Ming Chang, Xishan Zhang, Rui Zhang, Zhipeng Zhao, Guanhua He, Shaoli Liu
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 RNN-style backprop and TA-GRU to ConvGRU recurrent fusion (ablation) | RecurrentBEV · nuScenes val (ResNet50) | mAP: 0.371 -> 0.388 | Improved |
| Add backprop of historical BEV features alone (without TA-GRU) (ablation) | ConvGRU recurrent fusion · nuScenes val (ResNet50) | mAP: 0.371 -> 0.375 | No clear effect |
| Add TA-GRU inner grid transformation (ablation) | ConvGRU recurrent fusion · nuScenes val (ResNet50) | mAOE: 0.634 -> 0.596 | Improved |
| Add history backprop to TA-GRU (ablation) | RecurrentBEV · nuScenes val (ResNet50) | mAP: 0.378 -> 0.388 | Improved |
| Use 5x5 instead of 1x1 conv kernel in TA-GRU (ablation) | TA-GRU · nuScenes val (ResNet50) | mAP: 0.388 -> 0.403 | Improved |
| Train with 16 instead of 8 frames (ablation) | RecurrentBEV · nuScenes val (ResNet50) | mAP: 0.403 -> 0.402 | No clear effect |
| Increase training frames from 0 to 8 (ablation) | RecurrentBEV · nuScenes val (ResNet50) | mAP: 0.318 -> 0.403 | Improved |
| Use RecurrentBEV instead of parallel fusion SOLOFusion | RecurrentBEV · nuScenes val inference | latency: 96ms -> 39ms | Improved |
| Use RecurrentBEV instead of plain recurrent baseline | RecurrentBEV · nuScenes val inference | mAP: 0.396 -> 0.445 | Improved |
| Backprop through the whole network (including image encoder) across time (ablation) | RecurrentBEV training · Training memory | memory (MB): OOM at 2 frames vs 18,735 MB | Hurt |
Closest papers by meaning
- RaCFormer: Towards High-Quality 3D Object Detection via Query-based Radar-Camera Fusion (CVPR 2025 · similarity 0.88)
- IS-Fusion: Instance-Scene Collaborative Fusion for Multimodal 3D Object Detection (CVPR 2024 · similarity 0.88)
- Instance-Aware Multi-Camera 3D Object Detection with Structural Priors Mining and Self-Boosting Learning (AAAI 2024 · similarity 0.88)
- GeoBEV: Learning Geometric BEV Representation for Multi-view 3D Object Detection (AAAI 2025 · similarity 0.88)
- CRT-Fusion: Camera, Radar, Temporal Fusion Using Motion Information for 3D Object Detection (NeurIPS 2024 · similarity 0.88)
- RCBEVDet: Radar-camera Fusion in Bird's Eye View for 3D Object Detection (CVPR 2024 · 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.
