Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training

Official implementation of 'Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training'.

The paper has been accepted by NeurIPS 2022.

Introduction

Point-M2AE is a strong Multi-scale MAE pre-training framework for hierarchical self-supervised learning of 3D point clouds. Unlike the standard transformer in MAE, we modify the encoder and decoder into pyramid architectures to progressively model spatial geometries and capture both fine-grained and high-level semantics of 3D shapes. We design a multi-scale masking strategy to generate consistent visible regions across scales, and reconstruct the masked coordinates from a global-to-local perspective.

Point-M2AE Models

The pre-trained Point-M2AE is evaluated by Linear SVM on both ModelNet40 and ScanObjectNN datasets, without downstream fine-tuning:

Task	Dataset	Config	MN40 Acc.	SONN Acc.	Ckpts	Logs
Pre-training	ShapeNet	point-m2ae.yaml	92.87%	72.07%	best-ckpt.pth	log

After pre-training, we fine-tune Point-M2AE on three downstream tasks:

Coming in a few days.

Task	Dataset	Config	Acc.	Ckpts	Logs
Classification	ModelNet40 (1k)	-	-	-	-
Classification	ScanObjectNN	-	-	-	-
Segmentation	ShapeNetPart	-	-	-	-

Task	Dataset	Config	5w10s	5w20s	10w10s	10w20s
Few-shot Cls.	ModelNet40	-	-	-	-	-

Requirements

Installation

Create a conda environment and install basic dependencies:

git clone https://github.com/ZrrSkywalker/Point-M2AE.git
cd Point-M2AE

conda create -n pointm2ae python=3.7
conda activate pointm2ae

pip install -r requirements.txt

# Install the according versions of torch and torchvision
conda install pytorch torchvision cudatoolkit

Install GPU-related packages:

# Chamfer Distance and EMD
cd ./extensions/chamfer_dist
python setup.py install --user
cd ../emd
python setup.py install --user

# PointNet++
pip install "git+https://github.com/erikwijmans/Pointnet2_PyTorch.git#egg=pointnet2_ops&subdirectory=pointnet2_ops_lib"

# GPU kNN
pip install --upgrade https://github.com/unlimblue/KNN_CUDA/releases/download/0.2/KNN_CUDA-0.2-py3-none-any.whl

Datasets

For pre-training and fine-tuning, please follow DATASET.md to install ShapeNet, ModelNet40, ScanObjectNN, and ShapeNetPart datasets, referring to Point-BERT. Specially for Linear SVM evaluation, download the official ModelNet40 dataset and put the unzip folder under data/.

The final directory structure should be:

│Point-M2AE/
├──cfgs/
├──datasets/
├──data/
│   ├──ModelNet/
│   ├──ModelNetFewshot/
│   ├──modelnet40_ply_hdf5_2048/
│   ├──ScanObjectNN/
│   ├──ShapeNet55-34/
│   ├──shapenetcore_partanno_segmentation_benchmark_v0_normal/
├──...

Get Started

Pre-training

Point-M2AE is pre-trained on ShapeNet dataset with the config file cfgs/pre-training/point-m2ae.yaml. Run:

CUDA_VISIBLE_DEVICES=0 python main.py --config cfgs/pre-training/point-m2ae.yaml --exp_name pre-train

To evaluate the pre-trained Point-M2AE by Linear SVM on ModelNet40, create the folder ckpts/ and download the ckpt-best.pth into it. You will get 92.87% by running:

CUDA_VISIBLE_DEVICES=0 python main.py --config cfgs/pre-training/point-m2ae.yaml --exp_name test_svm --test_svm modelnet40 --ckpts ./ckpts/ckpt-best.pth

Fine-tuning

Coming in a few days.

Acknowledgement

This repo benefits from Point-BERT and Point-MAE. Thanks for their wonderful works.

Citation

@article{zhang2022point,
  title={Point-M2AE: Multi-scale Masked Autoencoders for Hierarchical Point Cloud Pre-training},
  author={Zhang, Renrui and Guo, Ziyu and Gao, Peng and Fang, Rongyao and Zhao, Bin and Wang, Dong and Qiao, Yu and Li, Hongsheng},
  journal={arXiv preprint arXiv:2205.14401},
  year={2022}
}

Contact

If you have any question about this project, please feel free to contact [email protected].