FASA: Feature Augmentation and Sampling Adaptation for Long-Tailed Instance Segmentation (ICCV 2021)

This repository contains the implementation of the following paper:

FASA: Feature Augmentation and Sampling Adaptation for Long-Tailed Instance Segmentation
Yuhang Zang,Chen Huang, Chen Change Loy
International Conference on Computer Vision (ICCV), 2021

[arXiv] [Project Page]

Running Environment

This code is based on mmdetection==2.14.0 and mmcv==1.3.9

Installation

Install mmdetection following the official instruction.
Install COCOAPI.

pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'

Init data directory:

mkdir data

Download LVIS data:

|-- data
`-- |-- lvis_v1
    `-- |-- annotations
        |   |-- lvis_v1_train.json
        |   `-- lvis_v1_val.json
        `-- images
            |-- train2017
            |   |-- 000000000009.jpg
            |   |-- 000000000025.jpg
            |   |-- ... 
            `-- val2017
                |-- 000000000139.jpg
                |-- 000000000285.jpg
                |-- ...

Train

./slurm_train.sh <config_file> <work_dir>

Evaluation

./slurm_test.sh <config_file> <checkpoint_path>

Results and models of LVIS v1

Backbone	Lr schd	Sampler	FASA	mask AP	mask APr	mask APc	mask APf	Config	Download
R-50-FPN	24e	Random	×	18.8	1.2	16.3	29.2	config	Google Drive
R-50-FPN	24e	Random	✓	22.2	10.5	20.4	29.4	config	Google Drive

Citation

If you find our work useful for your research, please consider citing the paper

@inproceedings{zang2021fasa,
  title={FASA: Feature Augmentation and Sampling Adaptation for Long-Tailed Instance Segmentation},
  author={Zang, Yuhang and Huang, Chen and Loy, Chen Change},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year={2021}
}

Contact

If you have any questions, please feel free to contact zang0012 AT ntu.edu.sg

License

This project is open sourced under MIT license.

yuhangzang/FASA