Weakly Supervised 3D Object Detection via Multi-Level Visual Guidance(ECCV2024)
Kuan-Chih Huang, Yi-Hsuan Tsai, Ming-Hsuan Yang
2DBoxのラベルのみを用いて3D検出器を学習させるVG-W3Dを提案.
画像とLiDAR点群で共通する情報を活用し,KITTI 3D Object Detectionにおいて,500フレームの3DBoxラベルを要する手法と同程度の精度を達成.
p2,6,7 FGR: Frustum-Aware Geometric Reasoning for Weakly Supervised 3D Vehicle Detection
p4 Weakly Supervised 3D Object Detection from Lidar Point Cloud
p5 MAP-Gen: An Automated 3D-Box Annotation Flow with Multimodal Attention Point Generator
p11 Emerging Properties in Self-Supervised Vision Transformers