First create a conda environment using the following command:

conda env create -f environment.yml

Then activate the environment using the following command:

conda activate chimera

Then download the required data and models from this link

In the recapture-detection directory, run the following command:

python main.py --test --test_path TEST_PATH --config CONFIG --test_raw_dirnames RAW_DIRNAMES --test_recap_dirnames RECAP_DIRNAMES

where TEST_PATH is the path to the model, CONFIG is the path to the configuration file, RAW_DIRNAMES is the list of raw directory names, and RECAP_DIRNAMES is the list of recapture directory names. Ensure that the configuration file matches the model being tested. Results should be printed to the terminal.

In the deepfake-detection directory, first update dataset paths in dataset_paths.py and move the deepfake detection model weights are in the pretrained_weights folder.

The following command:

./test.sh

will run all the deepfake detection tests at once and save the results in the deepfake-detection/results directory.

To train Chimera, you must collect data using a fixed screen and camera setup. Then, use the pix2pix training script to train the model. Training is done in the pytorch-CycleGAN-and-pix2pix directory.

Examples of training commands are provided in the pytorch-CycleGAN-and-pix2pix/scripts/train_pix2pix.sh file; ideal parameters depend on setup. Refer to the paper for more details on the training process.