https://arxiv.org/abs/2203.12119

Please refer to the env_setup.sh script for setting up the environment. This script is borrowed from the VPT project and can be found here: VPT GitHub.

The env_setup.sh script will help you configure the necessary dependencies and environment settings required to run the project smoothly.

-To fine-tune the model with your own data, you can refer to the data usage in the VPT project. For more details, visit the VPT GitHub repository: VPT GitHub.

-To preprocess your data, use the docDatasetSplit.py script located in the dataprepare folder. This script helps you split and prepare your data for subsequent model fine-tuning and training.

To prepare for training, you need to download and place the pre-trained Transformer-based backbones into the MODEL.MODEL_ROOT directory.

1. Download Pre-trained Models:

   • Download the backbone pre-trained models from the provided URL (VPT backbone models).

2. Rename Pre-trained ViT Model:

   • You also need to rename the downloaded ViT-B/16 checkpoint file from ViT-B_16.npz to imagenet21k_ViT-B_16.npz.

Ensure that all files are placed correctly and named appropriately for the model to load successfully.

For convenience, you can use the scripts in the TrainOrTestCmdShell folder for training and testing the model:

• Training: Use the dataTrain.sh script to train the model.
• Testing: Use the CrossDatatest.sh script to test the model.

These scripts simplify the process of running training and testing commands, allowing you to focus on the experimentation and evaluation of your models.

If you find our work helpful in your research, please cite it as:

@inproceedings{chen2024cma,
  title={CMA: A Chromaticity Map Adapter for Robust Detection of Screen-Recapture Document Images},
  author={Chen, Changsheng and Lin, Liangwei and Chen, Yongqi and Li, Bin and Zeng, Jishen and Huang, Jiwu},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={15577--15586},
  year={2024}
}

We would like to express our sincere gratitude to the authors of the VPT project for their invaluable contributions. The open-source code provided by the VPT team has been instrumental in the development of this project.

Special thanks to the VPT team for their dedication to advancing research and providing a robust foundation for others to build upon. You can find their work here: VPT GitHub.