Awesome
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<br> Yuhang Zang,Chen Huang, Chen Change Loy<br> International Conference on Computer Vision (ICCV), 2021<br>
[arXiv] [Project Page]
<p align="center"> <img width=95% src="./asserts/framework.png"> </p>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.