View Code? Open in Web Editor NEW

Implementation of joint construction of Mask/BBox heads in QDTrack-mots for joint training research

License: Apache License 2.0

Python 99.68% Makefile 0.03% Batchfile 0.04% Shell 0.26%

joint-qdtrack's Introduction

Implementations of Video Instance Segmentation methods on amodal video datasets

Datasets:

SAIL-VOS & SAIL-VOScut (videos split into video-cuts without abrupt scene change)

amodal annotations: visible annotations:

Amodal & Visible: QDTrack-mots-joint(+)

Using joint construction of the functional heads (Mask Heads / BBox Heads) in the original Mask R-CNN architecture of QDTrack-mots for joint training research.

QDTrack-mots-joint testing results:

amodal results: visible results:

QDTrack-mots-joint+ testing results:

amodal results: visible results:

Amodal / Visible:

Please refer to QDTrack for details of (Amodal)QDTrack-mots
Please refer to PCAN for details of (Amodal)PCAN

References

@inproceedings{hu2019sail,
  title={Sail-vos: Semantic amodal instance level video object segmentation-a synthetic dataset and baselines},
  author={Hu, Yuan-Ting and Chen, Hong-Shuo and Hui, Kexin and Huang, Jia-Bin and Schwing, Alexander G},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={3105--3115},
  year={2019}
}
@inproceedings{pang2021quasi,
  title={Quasi-dense similarity learning for multiple object tracking},
  author={Pang, Jiangmiao and Qiu, Linlu and Li, Xia and Chen, Haofeng and Li, Qi and Darrell, Trevor and Yu, Fisher},
  booktitle={Proceedings of the IEEE/CVF conference on computer vision and pattern recognition},
  pages={164--173},
  year={2021}
}
@inproceedings{pcan,
  title={Prototypical Cross-Attention Networks for Multiple Object Tracking and Segmentation},
  author={Ke, Lei and Li, Xia and Danelljan, Martin and Tai, Yu-Wing and Tang, Chi-Keung and Yu, Fisher},
  booktitle={Advances in Neural Information Processing Systems},
  year={2021}
}

Recommend Projects