BEVFix& Co-Fix3d & Co-Fusion & Co-Stream
welcome to replicate my experiment if you have GPUs available. I used 4 NVIDIA 4090 GPUs, and I estimate that using 8 GPUs would yield even better results.
LiDAR-Based 3D Object detection 1st on nuScenes data
Co-Fix3d mutil-mode rank 55th on the leader
Co-Fusion mutil-mode rank 25th on the leader
| nuScenes test dataset | Method | Modality | mAP | NDS |
|---|---|---|---|---|
| BEVFix | L | 68.0 | 71.9 | |
| Co-Fix3D | L | 69.1 | 72.9 | |
| Co-Fix3D | L+C | 72.3 | 74.1 | |
| Co-Fusion | L+C | 74.1 | 75.3 |
| nuScenes val dataset | Method | Modality | mAP | NDS |
|---|---|---|---|---|
| BEVFix | L | 65.5 | 70.0 | |
| Co-Fix3D | L | 66.8 | 71.3 | |
| Co-Fix3D | L+C | 70.8 | 73.1 | |
| Co-Fusion | L+C | 73.1 | 74.2 |
ToDo Multi-Frame Fusion: Co-Stream is coming soon