This repository is the official implementation of SWEM: Towards Real-Time Video Object Segmentation with Sequential Weighted Expectation-Maximization (CVPR2022)

• We use one NVIDIA V100 (16 GB Memory), whereas two 1080ti GPUs are also satisfied. Note that if you use one 1080ti, you can reduce the batch size and increase number of iterations correspondingly.

• To install requirements, run:

pip3 install -r requirements.txt

• Image Data: Download and process image datasets from STCN or directly download from Google Drive

• Video Data: Download the DAVIS17-TrainVal and YouTube-VOS datasets for main-training and testing.

• Modify the data path in configs/config.py

• Main training and testing

sh train_swem_s3.sh

This repository is released for academic use only. If you want to use our codes for commercial products, please contact linchrist@163.com in advance.

https://github.com/seoungwugoh/STM

https://github.com/haochenheheda/Training-Code-of-STM

https://github.com/hkchengrex/STCN

Codes of data samplers are from https://github.com/dvlab-research/Simple-SR

  @inproceedings{SWEM,
  title={SWEM: Towards Real-Time Video Object Segmentation with Sequential Weighted Expectation-Maximization},
  author={Lin, Zhihui and Yang, Tianyu and Li, Maomao and Wang, Ziyu and Yuan, Chun and Jiang, Wenhao and Liu, Wei},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={1362--1372},
  year={2022}
  }