Awesome
Instance As Identity: A Generic Online Paradigm for Video Instance Segmentation
Introduction
This repo is the official implementation of IAI paradigm for video instance segmentation. The overview of IAI paradigm is shown below.
<img src='doc/IAI_framework.png'>Visualization on YTVIS-2019 & OVIS
<img src='doc/visualization1.gif' width="400"> <img src='doc/visualization2.gif' width="400">
Performance
| Dataset | Model Name | Backbone | mAP | FPS | FPS (4GPUs) |
|---|---|---|---|---|---|
| YTVIS2019 | IAI+CondInst | R50 | 39.9 | 32 | 101 |
| YTVIS2019 | IAI+CondInst | R101 | 43.7 | 28 | 96 |
| YTVIS2021 | IAI+CondInst | R50 | 38.0 | 30 | 109 |
| OVIS | IAI+CondInst | R50 | 20.6 | 14 | 50 |
Installation
Please refer to INSTALL.md for installation instructions.
Getting Started
Please refer to Preparing Dataset.
Please refer to Getting Started with IAI.
Citation
@inproceedings{iai,
title={Instance As Identity: A Generic Online Paradigm for Video Instance Segmentation},
author={Zhu, Feng and Yang, Zongxin and Yu, Xin and Yang, Yi and Wei, Yunchao},
booktitle={ECCV},
year={2022},
}
Acknowledegement
This repo is based on MMDetection and AOT. Thanks for their wonderful works.