Exploring Geometry-aware Contrast and Clustering Harmonization for Self-supervised 3D Object Detection