Unsupervised Point Cloud Pre-Training via Occlusion Completion
This work addresses the challenge of effective pre-training for point cloud data, which is incremental as it builds on existing pre-training approaches with a novel masking strategy.
The paper tackles the problem of unsupervised pre-training for point clouds by introducing a method that masks occluded points and learns to reconstruct them, which improves accuracy across various downstream tasks like object classification and segmentation, outperforming previous methods.
We describe a simple pre-training approach for point clouds. It works in three steps: 1. Mask all points occluded in a camera view; 2. Learn an encoder-decoder model to reconstruct the occluded points; 3. Use the encoder weights as initialisation for downstream point cloud tasks. We find that even when we construct a single pre-training dataset (from ModelNet40), this pre-training method improves accuracy across different datasets and encoders, on a wide range of downstream tasks. Specifically, we show that our method outperforms previous pre-training methods in object classification, and both part-based and semantic segmentation tasks. We study the pre-trained features and find that they lead to wide downstream minima, have high transformation invariance, and have activations that are highly correlated with part labels. Code and data are available at: https://github.com/hansen7/OcCo