README.md

SAM-TTT Overview

Overview of our SAM-TTT framework: the Reverse SAM Parameter Configuration Module (R-SAMPC) and the T-vision Module (TVM). In the parallel phase, R-SAMPC and TVM operate independently, while in the fusion phase, the effectiveness of both modules is integrated.

Experiment Results

Qualitative results.

Quantitative results.

Weights

• SAM-TTT: https://drive.google.com/file/d/1FffLFfBre1PXXU19NO97JfXR5k9Fxs70/view?usp=drive_link

The predicted image

• SAM-TTT: https://drive.google.com/file/d/110032Gh0VSUSfl5d2W2oaHVa9HB1OuVq/view?usp=drive_link

Experiment Setting

The YML file is for reference only. There are redundant environments and not all of them need to be installed.

dataset

Refer to [COMPrompter]

Training

For the training process, run:

python My_Train.py

Testing / Inference

And run:

python Inference.py

Citation

If you find this project useful, please consider citing:

@inproceedings{yu2025sam,
  title={SAM-TTT: Segment Anything Model via Reverse Parameter Configuration and Test-Time Training for Camouflaged Object Detection},
  author={Yu, Zhenni and Zhao, Li and Xiao, Guobao and Zhang, Xiaoqin},
  booktitle={Proceedings of the 33rd ACM International Conference on Multimedia},
  pages={4030--4038},
  year={2025}
}