PETRv2: A Unified Framework for 3D Perception from Multi-Camera Images

Benchmark Model Rank Results
3d-lane-detection-on-openlanePETRv2-V∗ (VoVNetV2 with 400 anchor points)#3F1 (all): 61.2
3d-lane-detection-on-openlanePETRv2-V (VoVNetV2)#4F1 (all): 57.8
3d-lane-detection-on-openlanePETRv2-E (EfficientNet)#9F1 (all): 51.9
bird-s-eye-view-semantic-segmentation-on-nuscenesPETRv2#11IoU lane - 224x480 - 100x100 at 0.5: 44.8
3d-object-detection-on-nuscenes-camera-onlyPETRv2-pure#19NDS: 59.2Future Frame: false
3d-object-detection-on-nuscenesPETRv2 (V2-99, multi-scale) (extra data)#43NDS: 0.591mAP: 0.508