Skip to content

YilmazKadir/Mask4Former

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Mask4Former: Mask Transformer for 4D Panoptic Segmentation (Renamed from MASK4D)

Kadir Yilmaz, Jonas Schult, Alexey Nekrasov, Bastian Leibe

RWTH Aachen University

Mask4Former is a transformer-based model for 4D Panoptic Segmentation, achieving a new state-of-the-art performance on the SemanticKITTI test set.

PyTorch Lightning Config: Hydra Code style: black

teaser



[Project Webpage] [arXiv]

News

  • 2023-01-29: Mask4Former accepted to ICRA 2024

  • 2023-09-28: Mask4Former on arXiv

Dependencies

The main dependencies of the project are the following:

python: 3.8
cuda: 11.7

You can set up a conda environment as follows

conda create --name mask4former python=3.8
conda activate mask4former

pip install torch==1.13.0+cu117 torchvision==0.14.0+cu117 --extra-index-url https://download.pytorch.org/whl/cu117

pip install -r requirements.txt --no-deps

pip install git+https://github.com/NVIDIA/MinkowskiEngine.git -v --no-deps

pip install git+https://github.com/facebookresearch/[email protected] --no-deps

Data preprocessing

After installing the dependencies, we preprocess the SemanticKITTI dataset.

python -m datasets.preprocessing.semantic_kitti_preprocessing preprocess \
--data_dir "PATH_TO_RAW_SEMKITTI_DATASET" \
--save_dir "data/semantic_kitti"

python -m datasets.preprocessing.semantic_kitti_preprocessing make_instance_database \
--data_dir "PATH_TO_RAW_SEMKITTI_DATASET" \
--save_dir "data/semantic_kitti"

Training and testing

Train Mask4Former:

python main_panoptic.py

In the simplest case the inference command looks as follows:

python main_panoptic.py \
general.mode="validate" \
general.ckpt_path='PATH_TO_CHECKPOINT.ckpt'

Or you can use DBSCAN to boost the scores even further:

python main_panoptic.py \
general.mode="validate" \
general.ckpt_path='PATH_TO_CHECKPOINT.ckpt' \
general.dbscan_eps=1.0

Trained checkpoint

Mask4Former

The provided model, trained after the submission, achieves 71.1 LSTQ without DBSCAN and 71.5 with DBSCAN post-processing.

BibTeX

@inproceedings{yilmaz24mask4former,
  title     = {{Mask4Former: Mask Transformer for 4D Panoptic Segmentation}},
  author    = {Yilmaz, Kadir and Schult, Jonas and Nekrasov, Alexey and Leibe, Bastian},
  booktitle = {{International Conference on Robotics and Automation (ICRA)}},
  year      = {2024}
}