[ECCV 2024] Strike a Balance in Continual Panoptic Segmentation

This is an official implementation of the paper "Strike a Balance in Continual Panoptic Segmentation", accepted by ECCV 2024. 📝 Paper 🤗 Hugging Face

Installation

Environment

To install the required environment (for CUDA 11.3), run:

conda create -n balconpas python=3.8.17
conda activate balconpas
conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=11.3 -c pytorch -c conda-forge
python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html
pip install git+https://github.com/cocodataset/panopticapi.git
pip install -r requirements.txt

CUDA kernel for MSDeformAttn

After preparing the required environment, run the following command to compile CUDA kernel for MSDeformAttn:

CUDA_HOME must be defined and points to the directory of the installed CUDA toolkit.

cd mask2former/modeling/pixel_decoder/ops
sh make.sh

Data Preparation

Please download the ADE20K dataset and its instance annotation from here, then place the dataset in or create a symbolic link to the ./datasets directory. The structure of data path should be organized as follows:

ADEChallengeData2016/
  images/
  annotations/
  objectInfo150.txt
  sceneCategories.txt
  annotations_instance/
  annotations_detectron2/
  ade20k_panoptic_{train,val}.json
  ade20k_panoptic_{train,val}/
  ade20k_instance_{train,val}.json

The directory annotations_detectron2 is generated by running python datasets/prepare_ade20k_sem_seg.py. Then, run python datasets/prepare_ade20k_pan_seg.py to combine semantic and instance annotations for panoptic annotations and run python datasets/prepare_ade20k_ins_seg.py to extract instance annotations in COCO format.

To fit the requirements of continual segmentation tasks, run python continual/prepare_datasets.py to reorganize the annotations (reorganized annotations will be placed in ./json).

To generate replay samples for each step, run python continual/memory_generator/memory_selection_pan.py, python continual/memory_generator/memory_selection_sem.py, and python continual/memory_generator/memory_selection_inst.py for panoptic, semantic, and instance segmentation, respectively (replay samples will be placed in ./json/memory).

Example data preparation

# for Mask2Former
cd datasets
wget http://data.csail.mit.edu/places/ADEchallenge/ADEChallengeData2016.zip
unzip ADEChallengeData2016.zip
cd ADEChallengeData2016
wget http://sceneparsing.csail.mit.edu/data/ChallengeData2017/annotations_instance.tar
tar -xvf annotations_instance.tar
cd ../..
python datasets/prepare_ade20k_sem_seg.py
python datasets/prepare_ade20k_pan_seg.py
python datasets/prepare_ade20k_ins_seg.py

# for continual segmentation
python continual/prepare_datasets.py

# for memory
python continual/memory_generator/memory_selection_pan.py
python continual/memory_generator/memory_selection_sem.py
python continual/memory_generator/memory_selection_inst.py

Getting started

Training

Herein, we provide an example script for training the 100-10 continual panoptic segmentation of the ADE20K dataset:

python train_continual.py --config-file configs/ade20k/panoptic-segmentation/100-10.yaml \
CONT.TASK 1 SOLVER.BASE_LR 0.0001 SOLVER.MAX_ITER 160000 OUTPUT_DIR ./output/ps/100-10/step1

for t in 2 3 4 5 6; do
  python train_continual.py --config-file configs/ade20k/panoptic-segmentation/100-10.yaml \
  CONT.TASK ${t} SOLVER.BASE_LR 0.00005 SOLVER.MAX_ITER 10000 OUTPUT_DIR ./output/ps/100-10/step${t}
done

For more training scripts, please refer to the ./scripts directory.

Evaluation

To evaluate the trained model, add the argument --eval-only --resume to the command line. For example:

python train_continual.py --eval-only --resume --config-file configs/ade20k/panoptic-segmentation/100-10.yaml \
CONT.TASK 6 OUTPUT_DIR ./output/ps/100-10/step6

Model Zoo

We provide the pre-trained weights of our BalConpas-R. The models can be downloaded from the following links (rename the downloaded model to model_final.pth and place it in the corresponding output directory of the last step):

CPS	100-50 (2 steps)	100-10 (6 steps)	100-5 (11 steps)	50-50 (3 steps)
BalConpas-R	37.55	36.49	32.82	35.81

CSS	100-50 (2 steps)	100-10 (6 steps)	100-5 (11 steps)	50-50 (3 steps)
BalConpas-R	44.01	40.43	36.79	41.33

CIS	50-50 (2 steps)	50-10 (6 steps)	50-5 (11 steps)
BalConpas-R	24.89	23.34	21.23

Citation

@inproceedings{chen2024strike,
  title={Strike a Balance in Continual Panoptic Segmentation},
  author={Chen, Jinpeng and Cong, Runmin and Luo, Yuxuan and Ip, Horace Ho Shing and Kwong, Sam},
  booktitle={ECCV},
  year={2024}
}

Acknowledgements

• This code is based on Mask2Former.