y0d4a / deepfake-scanner Goto Github PK

This repository contains the command-line deepfake scanner tool with the pre-trained models that are currently used at deepware.ai.

To get an in-depth review about the scanner and the training process please refer to deepware.md. Here we will discuss the usage of the command-line scanner.

Make sure you have a nvidia gpu with cuda support. Scanner can run both on Windows and Linux.

• Clone the repo or download it as a zip and extract to a directory.
• Install the dependencies listed in requirements file with pip install -r requirements.txt
• Download the pre-trained model and place it in the weights directory.

We packed all the requirements in a portable zip file including the model. You can download the zip file and start scanning right away.

The scanning script is scan.py and it has four command line arguments. Here's the usage printed by the script.

scan.py <scan_dir> <models_dir> <cfg_file> <device>

Let's dive into the arguments.

• scan_dir is the directory of videos, alternatively a file with list of video paths is supported.
• models_dir is the directory pre-trained models are stored. There can be multiple models and they will be ensembled automatically.
• cfg_file is the config file. You shouldn't worry about this unless you want to train a new model.
• device is the cuda device that will be used for scanning. cpu is not supported as of now.

Here's an example command line.

scan.py /home/user/videos weights config.json cuda:0

On windows you can use scan.bat file with just the video folder as input.

Training scripts with a subset of DFDC dataset will be published soon. Stay tuned! 🔔