DeepFake Defenders

If you like our project, please give us a star ⭐ on GitHub for latest update.

[](https://github.com/VisionRush/DeepFakeDefenders/blob/main/LICENSE)  [](https://hits.seeyoufarm.com)  [](https://github.com/PKU-YuanGroup/MoE-LLaVA/issues?q=is%3Aopen+is%3Aissue) [](https://github.com/PKU-YuanGroup/MoE-LLaVA/issues?q=is%3Aissue+is%3Aclosed)

💡 We also provide [中文文档 / CHINESE DOC] and [한국어 문서 / KOREAN DOC]. We very welcome and appreciate your contributions to this project.

📣 News

• [2024.09.05] 🔥 We officially released the initial version of Deepfake defenders, and we won the third prize in the deepfake challenge at [the conference on the bund].

🚀 Quickly Start

1. Pretrained Models Preparation

Before getting started, please place the ImageNet-1K pretrained weight files in the ./pre_model directory. The download links for the weights are provided below:

RepLKNet: https://drive.google.com/file/d/1vo-P3XB6mRLUeDzmgv90dOu73uCeLfZN/view?usp=sharing
ConvNeXt: https://dl.fbaipublicfiles.com/convnext/convnext_base_1k_384.pth

2. Training from Scratch

2.1 Modifying the dataset path

Place the training-set (*.txt) file, validation-set (*.txt) file, and label (*.txt) file required for training in the dataset folder and name them with the same file name (there are various txt examples under dataset)

2.2 Modifying the Hyper-parameters

For the two models (RepLKNet and ConvNeXt) used, the following parameters need to be changed in main_train.py:

# For RepLKNet
cfg.network.name = 'replknet'; cfg.train.batch_size = 16
# For ConvNeXt
cfg.network.name = 'convnext'; cfg.train.batch_size = 24

2.3 Using the training script

Multi-GPUs:（8 GPUs were used）

bash main.sh

Single-GPU：

CUDA_VISIBLE_DEVICES=0 python main_train_single_gpu.py

2.4 Model Assembling

Replace the ConvNeXt trained model path and the RepLKNet trained model path in merge.py, and execute python merge.py to obtain the final inference test model.

2.5 Inference

The following example uses the POST request interface to request the image path as the request parameter, and the response output is the deepfake score predicted by the model.

#!/usr/bin/env python
# -*- coding:utf-8 -*-
import requests
import json
import requests
import json

header = {
    'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/92.0.4515.107 Safari/537.36'
}

url = 'http://ip:10005/inter_api'
image_path = './dataset/val_dataset/51aa9b8d0da890cd1d0c5029e3d89e3c.jpg'
data_map = {'img_path':image_path}
response = requests.post(url, data=json.dumps(data_map), headers=header)
content = response.content
print(json.loads(content))

3. Deploy in Docker

Building

sudo docker build  -t vision-rush-image:1.0.1 --network host .

Running

sudo docker run -d --name  vision_rush_image  --gpus=all  --net host  vision-rush-image:1.0.1

Star History