Point Cloud Pre-training
14 papers with code • 0 benchmarks • 0 datasets
Point cloud data represents 3D shapes as a set of discrete points in 3D space. This kind of data is primarily sourced from 3D scanners, LiDAR systems, and other similar technologies. Point cloud processing has a wide range of applications, such as robotics, autonomous vehicles, and augmented/virtual reality.
Pre-training on point cloud data is similar in spirit to pre-training on images or text. By pre-training a model on a large, diverse dataset, it learns essential features of the data type, which can then be fine-tuned on a smaller, task-specific dataset. This two-step process (pre-training and fine-tuning) often results in better performance, especially when the task-specific dataset is limited in size.
Benchmarks
These leaderboards are used to track progress in Point Cloud Pre-training
Most implemented papers
Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training
By fine-tuning on downstream tasks, Point-M2AE achieves 86. 43% accuracy on ScanObjectNN, +3. 36% to the second-best, and largely benefits the few-shot classification, part segmentation and 3D object detection with the hierarchical pre-training scheme.
PointContrast: Unsupervised Pre-training for 3D Point Cloud Understanding
facebookresearch/PointContrast
•
•
ECCV 2020
To this end, we select a suite of diverse datasets and tasks to measure the effect of unsupervised pre-training on a large source set of 3D scenes.
Unsupervised Point Cloud Pre-Training via Occlusion Completion
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.
Self-supervised Point Cloud Representation Learning via Separating Mixed Shapes
Albeit simple, the pre-trained encoder can capture the key points of an unseen point cloud and surpasses the encoder trained from scratch on downstream tasks.
Point Cloud Pre-Training With Natural 3D Structures
Moreover, the PC-FractalDB pre-trained model is especially effective in training with limited data.
POS-BERT: Point Cloud One-Stage BERT Pre-Training
fukexue/pos-bert
•
•
We propose to use the dynamically updated momentum encoder as the tokenizer, which is updated and outputs the dynamic supervision signal along with the training process.
ProposalContrast: Unsupervised Pre-training for LiDAR-based 3D Object Detection
Existing approaches for unsupervised point cloud pre-training are constrained to either scene-level or point/voxel-level instance discrimination.
Boosting Point-BERT by Multi-choice Tokens
Masked language modeling (MLM) has become one of the most successful self-supervised pre-training task.
BEV-MAE: Bird's Eye View Masked Autoencoders for Point Cloud Pre-training in Autonomous Driving Scenarios
vdigpku/bev-mae
•
•
Based on the property of outdoor point clouds in autonomous driving scenarios, i. e., the point clouds of distant objects are more sparse, we propose point density prediction to enable the 3D encoder to learn location information, which is essential for object detection.
PointClustering: Unsupervised Point Cloud Pre-Training Using Transformation Invariance in Clustering
fuchenustc/pointclustering
•
•
CVPR 2023
Feature invariance under different data transformations, i. e., transformation invariance, can be regarded as a type of self-supervision for representation learning.
