BFANet

[CVPR2025] BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis

Backbones

• Support OctFomer backbone;
• Support MinkowskiEngine backbone;
• Support PTv3 backbone;

Datasets

• Support ScanNetv2 Dataset;
• Support ScanNet200 Dataset;
• Support ScanNet++ Dataset;
• Support S3DIS Dataset;
• Support SemanticKITTI Dataset;

Others

• Release training and testing code of BFANet;
• Support TTA (Test Time Augmentation);
• Evaluation of Four Proposed Metrics

Environments

Our code was verified on Four RTX 4090 with CUDA 11.8 and Python 3.8.

Creat Conda Environment

conda create -n BFANet python=3.8
conda activate BFANet

Install OctFormer

conda install pytorch==1.12.1 torchvision torchaudio cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt
cd lib
git clone https://github.com/octree-nn/dwconv.git
pip install ./dwconv

Further OctFormer information can be found in OctFormer

Install Segmentator

cd segmentator
cd csrc && mkdir build && cd build
conda install cmake==3.26.4 cudnn

cmake .. \
-DCMAKE_PREFIX_PATH=`python -c 'import torch;print(torch.utils.cmake_prefix_path)'` \
-DPYTHON_INCLUDE_DIR=$(python -c "from distutils.sysconfig import get_python_inc; print(get_python_inc())")  \
-DPYTHON_LIBRARY=$(python -c "import distutils.sysconfig as sysconfig; print(sysconfig.get_config_var('LIBDIR'))") \
-DCMAKE_INSTALL_PREFIX=`python -c 'from distutils.sysconfig import get_python_lib; print(get_python_lib())'`

make && make install # after install, please do not delete this folder (as we only create a symbolic link)
cd ../../../../

Further segmentator information can be found in DKNet and Segmentator.

Install Our Pseudo-label Lib

cd lib/BFANet_lib
python setup.py develop
cd ../

Install Minkowski Engine

conda install openblas-devel -c anaconda
git clone https://github.com/NVIDIA/MinkowskiEngine.git
cd MinkowskiEngine
python setup.py install --blas_include_dirs=${CONDA_PREFIX}/include --blas=openblas
cd ../../

Further MinkowskiEngine information can be found in MinkowskiEngine

PointTransformerV3 environment

pip install spconv-cu118
conda install yapf addict einops scipy plyfile termcolor timm -c conda-forge -y
conda install pytorch-cluster pytorch-scatter pytorch-sparse -c pyg -y

Further PointTransformerV3 information can be found in PointTransformerV3

Dataset Preparation

(1) Download the ScanNet dataset.

(2) Put the data in the corresponding folders. The dataset files are organized as follows.

• Copy the files [scene_id]_vh_clean_2.ply, [scene_id]_vh_clean_2.0.010000.segs.json, [scene_id].aggregation.json and [scene_id]_vh_clean_2.labels.ply into the datasets/scannetv2/train and dataset/scannetv2/val folders according to the ScanNet v2 train/val split.

• Copy the files [scene_id]_vh_clean_2.ply into the datasets/scannetv2/test folder according to the ScanNet v2 test split.

• Put the file scannetv2-labels.combined.tsv in the datasets/scannetv2 folder.

BFANet
├── data
│   ├── ScanNet
│   │   ├── train
│   │   │   ├── [scene_id]_vh_clean_2.ply & [scene_id]_vh_clean_2.0.010000.segs.json & [scene_id].aggregation.json & [scene_id]_vh_clean_2.labels.ply
│   │   ├── val
│   │   │   ├── [scene_id]_vh_clean_2.ply & [scene_id]_vh_clean_2.0.010000.segs.json & [scene_id].aggregation.json & [scene_id]_vh_clean_2.labels.ply
│   │   ├── test
│   │   │   ├── [scene_id]_vh_clean_2.ply 
│   │   ├── scannetv2-labels.combined.tsv

(3) Decode the files to the "BFANet/datasets/ScanNetv2/npy/", if you don't want to use shared memory, please set the "use_shm_flag" as False in "BFANet/datasets/ScanNetv2/data_decode.py"

cd BFANet
export PYTHONPATH=./
python datasets/ScanNetv2/data_decode.py

Environments

cd BFANet
python train.py

Citation

If you find this work useful in your research, please cite:

@inproceedings{zhao2025bfanet,
  title={BFANet: Revisiting 3D Semantic Segmentation with Boundary Feature Analysis},
  author={Zhao, Weiguang and Zhang, Rui and Wang, Qiufeng and Cheng, Guangliang and Huang, Kaizhu},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR)},
  pages={29395--29405},
  year={2025}
}

Acknowlegement

This project is not possible without multiple great opensourced codebases. We list some notable examples: OctFormer, Seg-Aliasing, PointTransformerV3,MinkowskiEngine, Superpoint Transformer, Mix3d, DKNet, etc.