Siamese Network with Interactive Transformer for Video Object Segmentation

This repository is the code release of the paper Siamese Network with Interactive Transformer for Video Object Segmentation, accepted by AAAI 2022.

Requirements

• Linux
• Python >= 3.6
• Pytorch >=1.5
• CUDA>=9.0
• Pillow, opencv-python, scipy

Training

Stage 1

Pretraining on MS-COCO.

python train_coco.py -Dcoco "path to coco" 

Stage 2

Training on Davis & Youtube-VOS.

python train_davis.py -Ddavis "path to davis" -Dyoutube "path to youtube-vos" -resume "path to coco pretrained weights"

Evaluation

Evaluating on DAVIS val set.

python eval.py -p "path to weights"

Demo

Acknowledgement

This codebase borrows the code and structure from official STM repository

Citing SITVOS

@article{lan2021siamese,
  title={Siamese Network with Interactive Transformer for Video Object Segmentation},
  author={Lan, Meng and Zhang, Jing and He, Fengxiang and Zhang, Lefei},
  journal={arXiv preprint arXiv:2112.13983},
  year={2021}
}