Sonata: Self-Supervised Learning of Reliable Point Representations
This work addresses the challenge of developing reliable self-supervised learning methods for 3D point cloud data, which is crucial for applications in robotics and autonomous systems, representing a strong specific gain rather than an incremental improvement.
The paper tackles the problem of unreliable self-supervised point cloud models for 3D tasks by addressing the 'geometric shortcut' issue, achieving a tripling of linear probing accuracy from 21.8% to 72.5% on ScanNet and near-doubling performance with only 1% of data.
In this paper, we question whether we have a reliable self-supervised point cloud model that can be used for diverse 3D tasks via simple linear probing, even with limited data and minimal computation. We find that existing 3D self-supervised learning approaches fall short when evaluated on representation quality through linear probing. We hypothesize that this is due to what we term the "geometric shortcut", which causes representations to collapse to low-level spatial features. This challenge is unique to 3D and arises from the sparse nature of point cloud data. We address it through two key strategies: obscuring spatial information and enhancing the reliance on input features, ultimately composing a Sonata of 140k point clouds through self-distillation. Sonata is simple and intuitive, yet its learned representations are strong and reliable: zero-shot visualizations demonstrate semantic grouping, alongside strong spatial reasoning through nearest-neighbor relationships. Sonata demonstrates exceptional parameter and data efficiency, tripling linear probing accuracy (from 21.8% to 72.5%) on ScanNet and nearly doubling performance with only 1% of the data compared to previous approaches. Full fine-tuning further advances SOTA across both 3D indoor and outdoor perception tasks.