3D Object Detection on KITTI (val) (Accuracy and Cost Summary)
29.5APProxySelect
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ProxySelectCandidate pool=One 3D-MOOD box and nine perturbations of position, dimensions, yaw, and class, Selector output=one selected 3D box, Fine-tuning data=3,000 KITTI images2026.06 | 29.5 | — | 1.5 | |
| SpecialistCandidate pool=One 3D-MOOD box and nine perturbations of position, dimensions, yaw, and class2026.06 | 22.2 | — | 1 | |
| Direct VLMDescription=Direct VLM regression2026.06 | 14.8 | — | 5.1 |