APES: Attention-based Point Cloud Edge Sampling

🔥 News

[Feb 2025] We have a follow-up paper Samble accepted by CVPR 2025!
[Mar 2023] APES is selected as a Highlight by CVPR 2023!
[Feb 2023] APES is accepted by CVPR 2023!

🏡 Homepage

This project is selected as a Highlight by CVPR 2023! For more information about the project, please refer to our project homepage.

🔧 Prerequisites

Install all necessary packages using:

conda create -n APES python=3.9 -y
conda activate APES
conda install pytorch==2.0.0 pytorch-cuda=11.7 -c pytorch -c nvidia -y
pip install -r requirements.txt

📚 Data

Download and preprocess the data using:

python utils/download_modelnet.py  # for classification
python utils/download_shapenet.py  # for segmentation

📦 Train

Train models from scratch using:

# using single GPU
# command: bash utils/single_gpu_train.sh cfg_file
bash utils/single_gpu_train.sh configs/apes/apes_cls_local-modelnet-200epochs.py  # for classification using local-based downsampling
bash utils/single_gpu_train.sh configs/apes/apes_cls_global-modelnet-200epochs.py  # for classification using global-based downsampling
bash utils/single_gpu_train.sh configs/apes/apes_seg_local-shapenet-200epochs.py  # for segmentation using local-based downsampling
bash utils/single_gpu_train.sh configs/apes/apes_seg_global-shapenet-200epochs.py  # for segmentation using global-based downsampling

# using multiple GPUs 
# command: bash utils/dist_train.sh cfg_file num_gpus
bash utils/dist_train.sh configs/apes/apes_cls_local-modelnet-200epochs.py 2  # for classification using local-based downsampling
bash utils/dist_train.sh configs/apes/apes_cls_global-modelnet-200epochs.py 2  # for classification using global-based downsampling
bash utils/dist_train.sh configs/apes/apes_seg_local-shapenet-200epochs.py 2  # for segmentation using local-based downsampling
bash utils/dist_train.sh configs/apes/apes_seg_global-shapenet-200epochs.py 2  # for segmentation using global-based downsampling

📦 Test

Test model with checkpoint using:

# using single GPU
# command: bash utils/single_gpu_test.sh cfg_file ckpt_path
bash utils/single_gpu_test.sh configs/apes/apes_cls_local-modelnet-200epochs.py ckpt_path  # for classification using local-based downsampling
bash utils/single_gpu_test.sh configs/apes/apes_cls_global-modelnet-200epochs.py ckpt_path # for classification using global-based downsampling
bash utils/single_gpu_test.sh configs/apes/apes_seg_local-shapenet-200epochs.py ckpt_path  # for segmentation using local-based downsampling
bash utils/single_gpu_test.sh configs/apes/apes_seg_global-shapenet-200epochs.py ckpt_path # for segmentation using global-based downsampling

# using multiple GPUs 
# command: bash utils/dist_test.sh cfg_file ckpt_path num_gpus
bash utils/dist_test.sh configs/apes/apes_cls_local-modelnet-200epochs.py ckpt_path 2  # for classification using local-based downsampling
bash utils/dist_test.sh configs/apes/apes_cls_global-modelnet-200epochs.py ckpt_path 2  # for classification using global-based downsampling
bash utils/dist_test.sh configs/apes/apes_seg_local-shapenet-200epochs.py ckpt_path 2  # for segmentation using local-based downsampling
bash utils/dist_test.sh configs/apes/apes_seg_global-shapenet-200epochs.py ckpt_path 2  # for segmentation using global-based downsampling

📦 Visualization

Visualize results with checkpoint using:

# using single GPU
# command: bash utils/single_gpu_test.sh cfg_file ckpt_path -vis
bash utils/single_gpu_test.sh configs/apes/apes_cls_local-modelnet-200epochs.py ckpt_path -vis  # for classification using local-based downsampling
bash utils/single_gpu_test.sh configs/apes/apes_cls_global-modelnet-200epochs.py ckpt_path -vis  # for classification using global-based downsampling
bash utils/single_gpu_test.sh configs/apes/apes_seg_local-shapenet-200epochs.py ckpt_path -vis  # for segmentation using local-based downsampling
bash utils/single_gpu_test.sh configs/apes/apes_seg_global-shapenet-200epochs.py ckpt_path -vis  # for segmentation using global-based downsampling

# using multiple GPUs 
# command: bash utils/dist_test.sh cfg_file ckpt_path num_gpus -vis
bash utils/dist_test.sh configs/apes/apes_cls_local-modelnet-200epochs.py ckpt_path 2 -vis  # for classification using local-based downsampling
bash utils/dist_test.sh configs/apes/apes_cls_global-modelnet-200epochs.py ckpt_path 2 -vis  # for classification using global-based downsampling
bash utils/dist_test.sh configs/apes/apes_seg_local-shapenet-200epochs.py ckpt_path 2 -vis  # for segmentation using local-based downsampling
bash utils/dist_test.sh configs/apes/apes_seg_global-shapenet-200epochs.py ckpt_path 2 -vis  # for segmentation using global-based downsampling

📖 Citation

If you are interested in this work, please cite as below:

@inproceedings{wu_2023_attention,
author={Wu, Chengzhi and Zheng, Junwei and Pfrommer, Julius and Beyerer, J\"urgen},
title={Attention-Based Point Cloud Edge Sampling},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2023}
}

@inproceedings{wu_2025_samble,
author={Wu, Chengzhi and Wan, Yuxin and Fu, Hao and Pfrommer, Julius and Zhong, Zeyun and Zheng, Junwei and Zhang, Jiaming and Beyerer, J\"urgen},
title={SAMBLE: Shape-Specific Point Cloud Sampling for an Optimal Trade-Off Between Local Detail and Global Uniformity},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2025}
}