DeAOT

DeAOT Inference Weights

Inference-ready DeAOT model weights with training-only checkpoint artifacts removed, resulting in substantially smaller files for inference and deployment.

Overview

These checkpoints are derived from the original DeAOT weights by removing training-only state and artifacts that are not required for inference.

No model architecture changes or retraining are performed. The resulting checkpoints retain the model state required for inference while being significantly smaller than the original training-oriented checkpoints.

Suitable for:

• Inference
• Deployment
• Containerized applications
• Production model distribution

Provenance

Derived from the original DeAOT model weights by z-x-yang.

• Upstream: DeAOT
• Paper: Decoupling Features in Hierarchical Propagation for Video Object Segmentation
• Original copyright: Copyright (c) 2020, z-x-yang
• Conference: NeurIPS 2022
• License: BSD 3-Clause License

License

Distributed under the BSD 3-Clause License.

See LICENSE for the complete license text.

Citation

If you use DeAOT, please cite the original work:

@article{yang2022deaot,
  title   = {Decoupling Features in Hierarchical Propagation for Video Object Segmentation},
  author  = {Yang, Zongxin and Yang, Yi},
  journal = {Advances in Neural Information Processing Systems},
  year    = {2022}
}

Acknowledgements

Thanks to z-x-yang and contributors for the original DeAOT implementation and model weights.