Point-BERT: Pre-Training 3D Point Cloud Transformers with Masked Point Modeling

Created by Xumin Yu*, Lulu Tang*, Yongming Rao*, Tiejun Huang, Jie Zhou, Jiwen Lu

[arXiv] [Project Page] [Models]

This repository contains PyTorch implementation for Point-BERT:Pre-Training 3D Point Cloud Transformers with Masked Point Modeling (CVPR 2022).

Point-BERT is a new paradigm for learning Transformers to generalize the concept of BERT onto 3D point cloud. Inspired by BERT, we devise a Masked Point Modeling (MPM) task to pre-train point cloud Transformers. Specifically, we first divide a point cloud into several local patches, and a point cloud Tokenizer is devised via a discrete Variational AutoEncoder (dVAE) to generate discrete point tokens containing meaningful local information. Then, we randomly mask some patches of input point clouds and feed them into the backbone Transformer. The pre-training objective is to recover the original point tokens at the masked locations under the supervision of point tokens obtained by the Tokenizer.

Pretrained Models

model	dataset	config	url
dVAE	ShapeNet	config	Tsinghua Cloud / BaiDuYun(code:26d3)
Point-BERT	ShapeNet	config	Tsinghua Cloud / BaiDuYun(code:jvtg)

model	dataset	Acc.	Acc. (vote)	config	url
Transformer	ModelNet	92.67	93.24	config	Tsinghua Cloud / BaiDuYun(code:tqow)
Transformer	ModelNet	92.91	93.48	config	Tsinghua Cloud / BaiDuYun(code:tcin)
Transformer	ModelNet	93.19	93.76	config	Tsinghua Cloud / BaiDuYun(code:k343)
Transformer	ScanObjectNN	88.12	--	config	Tsinghua Cloud / BaiDuYun(code:f0km)
Transformer	ScanObjectNN	87.43	--	config	Tsinghua Cloud / BaiDuYun(code:k3cb)
Transformer	ScanObjectNN	83.07	--	config	Tsinghua Cloud / BaiDuYun(code:rxsw)

Usage

Requirements

• PyTorch >= 1.7.0
• python == 3.7
• CUDA >= 10.2
• GCC >= 4.9
• torchvision
• timm
• open3d
• tensorboardX

pip install -r requirements.txt

Building Pytorch Extensions for Chamfer Distance, PointNet++ and kNN

NOTE: PyTorch >= 1.7 and GCC >= 4.9 are required.

# Chamfer Distance
bash install.sh
# PointNet++
pip install "git+git://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

Dataset

We use ShapeNet for the training of dVAE and the pre-training of Point-BERT models. And finetuning the Point-BERT models on ModelNet, ScanObjectNN, ShapeNetPart The details of used datasets can be found in DATASET.md.

dVAE

To train a dVAE by yourself, simply run:

bash scripts/train.sh <GPU_IDS>\
    --config cfgs/ShapeNet55_models/dvae.yaml \
    --exp_name <name>

Visualize the reconstruction results of a pre-trained dVAE, run: (default path: ./vis)

bash ./scripts/test.sh <GPU_IDS> \
    --ckpts <path>\
    --config cfgs/ShapeNet55_models/dvae.yaml\
    --exp_name <name>

Point-BERT pre-training

To pre-train the Point-BERT models on ShapeNet, simply run: (complete the ckpt in cfgs/Mixup_models/Point-BERT.yaml first )

bash ./scripts/dist_train_BERT.sh <NUM_GPU> <port>\
    --config cfgs/Mixup_models/Point-BERT.yaml \
    --exp_name pointBERT_pretrain 
    [--val_freq 10]

val_freq controls the frequence to evaluate the Transformer on ModelNet40 with LinearSVM.

Fine-tuning on downstream tasks

We finetune our Point-BERT on 4 downstream tasks: Classfication on ModelNet40, Few-shot learning on ModelNet40, Transfer learning on ScanObjectNN and Part segmentation on ShapeNetPart.

ModelNet40

To finetune a pre-trained Point-BERT model on ModelNet40, simply run:

# 1024 points
bash ./scripts/train_BERT.sh <GPU_IDS> \
    --config cfgs/ModelNet_models/PointTransformer.yaml\
    --finetune_model\
    --ckpts <path>\
    --exp_name <name>
# 4096 points
bash ./scripts/train_BERT.sh <GPU_IDS>\
    --config cfgs/ModelNet_models/PointTransformer_4096point.yaml\ 
    --finetune_model\ 
    --ckpts <path>\
    --exp_name <name>
# 8192 points
bash ./scripts/train_BERT.sh <GPU_IDS>\
    --config cfgs/ModelNet_models/PointTransformer_8192point.yaml\ 
    --finetune_model\ 
    --ckpts <path>\
    --exp_name <name>

To evaluate a model finetuned on ModelNet40, simply run:

bash ./scripts/test_BERT.sh <GPU_IDS>\
    --config cfgs/ModelNet_models/PointTransformer.yaml \
    --ckpts <path> \
    --exp_name <name>

Few-shot Learning on ModelNet40

We follow the few-shot setting in the previous work.

First, generate your own few-shot learning split or use the same split as us (see DATASET.md).

# generate few-shot learning split
cd datasets/
python generate_few_shot_data.py
# train and evaluate the Point-BERT
bash ./scripts/train_BERT.sh <GPU_IDS> \
    --config cfgs/Fewshot_models/PointTransformer.yaml \
    --finetune_model \
    --ckpts <path> \
    --exp_name <name> \
    --way <int> \
    --shot <int> \
    --fold <int>

ScanObjectNN

To finetune a pre-trained Point-BERT model on ScanObjectNN, simply run:

bash ./scripts/train_BERT.sh <GPU_IDS>  \
    --config cfgs/ScanObjectNN_models/PointTransformer_hardest.yaml \
    --finetune_model \
    --ckpts <path> \
    --exp_name <name>

To evaluate a model on ScanObjectNN, simply run:

bash ./scripts/test_BERT.sh <GPU_IDS>\
    --config cfgs/ScanObjectNN_models/PointTransformer_hardest.yaml \
    --ckpts <path> \
    --exp_name <name>

Part Segmentation

To finetune a pre-trained Point-BERT model on ShapeNetPart

cd segmentation
python train_partseg.py \
    --model PointTransformer \
    --gpu <GPU_IDS> \
    --pretrain_weight <path> \
    --log_dir <name> 

To evaluate a model on ShapeNetPart, simply run:

python test_partseg.py \
    --gpu <GPU_IDS> \
    --log_dir <name> 

Visualization

Masked point clouds reconstruction using our Point-BERT model trained on ShapeNet

License

MIT License

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{yu2021pointbert,
  title={Point-BERT: Pre-Training 3D Point Cloud Transformers with Masked Point Modeling},
  author={Yu, Xumin and Tang, Lulu and Rao, Yongming and Huang, Tiejun and Zhou, Jie and Lu, Jiwen},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2022}
}