Implementation of CVPR2021 oral paper (best paper candidate), "Guided Interactive Video Object Segmentation Using Reliability-Based Attention Maps"

[Paper] [Project page] [Video-introduction] [Video-comparison]

Codes in this github:

1. DAVIS2017 evaluation based on the DAVIS framework
2. Youtube-iVOS evaluation based on the Youtube2019
3. DAVIS2017 real-world evaluation GUI - linked to GUI-iVOS_and_GIS

• cuda 11.0
• python 3.6
• pytorch 1.6.0
• davisinteractive 1.0.4
• numpy, cv2, PtQt5, and other general libraries of python3

• root/checkpoints: save our checkpoints (pth extensions) here.

• root/dataset_torch: pytorch datasets.

• root/libs: library of utility files.

• root/networks : codes for networks

  • deeplab: applies ASPP module in decoders. [original code]
  • network.py: consists our whole network.

• root/results : result files for both evaluation results are attached.

• root/config.py : configurations. you must set your directories here.

• root/IVOS_main_DAVIS.py : DAVIS2017 evaluation based on the DAVIS framework.

• root/IVOS_main_youtube.py : Youtube-iVOS evaluation based on the Youtube2019.

1. Edit config.py to set the directory of your DAVIS2017 dataset and the other configurations.

2. Download our parameters and place the file as root/checkpoints/GIS-ckpt_standard.pth.

   • For DAVIS2017 evaluation [Google-Drive]
   • For youtube2019 evaluation [Google-Drive]

3. Evaluate with

   • python3 IVOS_main_DAVIS.py.
   • python3 IVOS_main_youtube.py.

Multi-object GUI (for DAVIS2017) is available at our github page, [GUI-IVOS]

Please cite our paper if the implementations are useful in your work:

@Inproceedings{
Yuk2021GIS,
title={Guided Interactive Video Object Segmentation Using Reliability-Based Attention Maps},
author={Yuk Heo and Yeong Jun Koh and Chang-Su Kim},
booktitle={CVPR},
year={2021},
url={https://openaccess.thecvf.com/content/CVPR2021/papers/Heo_Guided_Interactive_Video_Object_Segmentation_Using_Reliability-Based_Attention_Maps_CVPR_2021_paper.pdf}
}