FastPoint

This is an implementation of our paper "FastPoint: Accelerating 3D Point Cloud Model Inference via Sample Point Distance Prediction".

Install

source install.sh

Note:

We recommend using CUDA 11.x; check your CUDA version by: nvcc --version before using the bash file;

Prepare Pretrained Models

Model weights used in our experiments are available through Google Drive link.

Prepare Dataset

S3DIS

mkdir -p data/S3DIS/
cd data/S3DIS
gdown https://drive.google.com/uc?id=1MX3ZCnwqyRztG1vFRiHkKTz68ZJeHS4Y
tar -xvf s3disfull.tar

ScanNet

cd data
gdown https://drive.google.com/uc?id=1uWlRPLXocqVbJxPvA2vcdQINaZzXf1z_
tar -xvf ScanNet.tar

SemanticKITTI

Download SemanticKITTI dataset.

Experiments

Inference Speed (val mode)

# Run baseline
CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/[s3dis, scannet, semantickitti]/[pointvector-l, pointmetabase-l].yaml wandb.use_wandb=False mode=val --pretrained_path [path]

# Run with only MDPS
CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/[s3dis, scannet, semantickitti]/[pointvector-l-mdps, pointmetabase-l-mdps].yaml wandb.use_wandb=False mode=val --pretrained_path [path]

# Run with FastPoint
CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/[s3dis, scannet, semantickitti]/[pointvector-l-fastpoint, pointmetabase-l-fastpoint].yaml wandb.use_wandb=False mode=val --pretrained_path [path]

Model Accuracy (test mode)

# Run baseline
CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/[s3dis, scannet, semantickitti]/[pointvector-l, pointmetabase-l].yaml wandb.use_wandb=False mode=test --pretrained_path [path]

# Run with only MDPS
CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/[s3dis, scannet, semantickitti]/[pointvector-l-mdps, pointmetabase-l-mdps].yaml wandb.use_wandb=False mode=test --pretrained_path [path]

# Run with FastPoint
CUDA_VISIBLE_DEVICES=0 bash script/main_segmentation.sh cfgs/[s3dis, scannet, semantickitti]/[pointvector-l-fastpoint, pointmetabase-l-fastpoint].yaml wandb.use_wandb=False mode=test --pretrained_path [path]

Training Minimum Distance Curve Estimator

We provide example training script for S3DIS estimator. Detailed hyperparameters for other datasets are explained in Appendix A.3.

CUDA_VISIBLE_DEVICES=0 bash script/train_estimator.sh cfgs/s3dis/pointmetabase-l.yaml wandb.use_wandb=False

Acknowledgment

This repository is built on reusing codes of Frugal_PN_Training, PointMetaBase, OpenPoints and PointNeXt.

Citation

@inproceedings {fastpoint,
    title={FastPoint: Accelerating 3D Point Cloud Model Inference via Sample Point Distance Prediction},
    author={Donghyun Lee and Dawoon Jeong and Jae W. Lee and Hongil Yoon},
    booktitle = {IEEE/CVF International Conference on Computer Vision ({ICCV} 25)},
    year={2025},
}