Unsupervised Point Cloud Pre-Training via Occlusion Completion
arXiv:2010.01089
Abstract
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
arXiv commentsync with ICCV camera ready
References in corpus (10)
- Efficient Estimation of Word Representations in Vector Space
- Learning Representations and Generative Models for 3D Point Clouds
- Understanding the Role of Individual Units in a Deep Neural Network
- Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds
- Strategies for Pre-training Graph Neural Networks
- PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding
- PointDAN: A Multi-Scale 3D Domain Adaption Network for Point Cloud Representation
- Self-Supervised Few-Shot Learning on Point Clouds
- Towards Semantic Segmentation of Urban-Scale 3D Point Clouds: A Dataset, Benchmarks and Challenges
- Self-Supervised Pretraining of 3D Features on any Point-Cloud