aim-uofa/SegPrompt

SegPrompt: Boosting Open-world Segmentation via Category-level Prompt Learning

Muzhi Zhu¹,   Hengtao Li¹,   Hao Chen¹,   Chengxiang Fan¹,   Weian Mao^2,1,   Chenchen Jing¹,   Yifan Liu²,   Chunhua Shen¹

¹Zhejiang University,   ²The University of Adelaide,  

News

• [2023/07/14] Our work SegPrompt is accepted by Int. Conf. Computer Vision (ICCV) 2023! 🎉🎉🎉
• [2023/08/30] We release our new benchmark LVIS-OW.

Installation

Please follow the instructions in Mask2Former

Other requirements

pip install torchshow
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.10.1+cu113.html
pip install lvis
pip install setuptools==59.5.0
pip install seaborn

LVIS-OW benchmark

Here we provide our proposed new benchmark LVIS-OW.

Dataset preparation

First prepare COCO and LVIS dataset, place them under $DETECTRON2_DATASETS following Detectron2

The dataset structure is as follows:

datasets/
  coco/
    annotations/
      instances_{train,val}2017.json
    {train,val}2017/
  lvis/
    lvis_v1_{train,val}.json

We reorganize the dataset and divide the categories into Known-Seen-Unseen to better evaluate the open-world model. The json files can be downloaded from here.

Or you can directly use the command to generate from the json file of COCO and LVIS.

bash tools/prepare_lvisow.sh 

After you successfully get lvis_v1_train_ow.json and lvis_v1_val_resplit_r.json, you can refer to here to register the training set and test set. Then you can use our benchmark for training and testing.

Evaluation on LVIS-OW

python tools/eval_lvis_ow.py --dt-json-file output/m2f_binary_lvis_ow/lvis_r/inference/lvis_instances_results.json

Acknowledgement

We thank the following repos for their great works:

• Mask2Former
• Detectron2

Cite our Paper

If you found this project useful for your paper, please kindly cite our paper.