IDNet-2025 is a synthetic data generation framework to generate a large number of identity documents using only a few documents from a target domain without including any private information.
Python>=10.0 is required to run the project. To install all the dependencies, either run the following command or manually install the dependencies in the requirements file.
pip install -r requirements.txtDownload the pretrained models from here and place the unzipped models folder inside the data/ directory.
Download the target domain images file from here and place the unzipped 'target_images' folder inside the data/ directory.
Download the datasets from here and place the unzipped folders inside the data/ directory.
In this Experiments, we used SIDTD tmplate dataset as our target domain and used in our experiments, you can download the SIDTD data here or here, then put the reals and 'fakes' folders inside the data/templates/ directory.
To run the Bayesian Optimization baseline, run the following command under the root directory of the project.
python experiments/Bayesian_search.py target_samples with_model lambda0 lambda1 candidate_modelsIn the above command, the parameter target_samples denotes the number of samples to be used (int), with_model denotes whether the optimization will be model-guided or not (0 or 1), lambda0 and lambda1 parameters control the fractions of similarity score and consistency score in the total evaluation score, and candidate_models stand for the names of models (space separated) that will guide the optimization. Example commands are given below:
python experiments/Bayesian_search.py 20 1 1 1 resnet50python experiments/Bayesian_search.py 20 0 1 1 ssimYou may also execute experiments/run.py which provides sample commands to execute Bayesian_search.py.
To run the Hyperband search baseline, run the following command under the root directory of the project.
python experiments/Hyperband_search.py param_r param_eta target_samples with_model lambda0 lambda1 candidate_modelsHere, param_r and param_eta represent maximum resources and successive halving parameters of the Hyperband search method. Other parameters target_samples, with_model, lambda0, lambda1, and candidate_models are similar to those in Bayesian search. Example commands:
python experiments/Hyperband_search.py 700 3 20 1 1 1 resnet50python experiments/Hyperband_search.py 700 3 20 0 1 1 ssimAlso, experiments/run_hyperband.py can be executed, which runs Hyperband_search.py with example commands.
Download the repo of CycleGAN here Download the data used for this baseline here
To run the CycleGAN baseline, run the following command under the root directory of the project.
cd experiments
bash cycle_run.shThis include the CycleGAN training part and results evaluation part(test_cyclegan.py)
see here for more details.
see here for more details.