Deep-Live-Cam Server [by Asimov Academy]

Este é um fork otimizado do Deep-Live-Cam original, focado em processamento distribuído via WebSocket para melhor performance em deepfakes em tempo real.

Diferencial desta Versão

Esta versão foi modificada para separar o processamento de imagens da interface do usuário, permitindo:

• Processamento remoto em servidores com GPUs potentes
• Menor latência na interface do usuário
• Possibilidade de múltiplos clientes conectados ao mesmo servidor
• Ideal para streaming e aplicações em tempo real

Arquitetura

Cliente (Webcam) <-> WebSocket <-> Servidor (GPU) 

• Cliente: Captura frames da webcam e exibe resultados
• WebSocket: Comunicação em tempo real de baixa latência
• Servidor: Processa as imagens usando GPU dedicada

Instalação

Requisitos do Sistema

• Python 3.10
• CUDA Toolkit 11.8.0 (para GPUs NVIDIA)
• ffmpeg

先用pip 安装 torch==2.6.0 torchvision==0.21.0,再安装 basicsr==1.4.2

Configuração do Ambiente

1. Clone o repositório:

git clone https://github.com/asimov-academy/Deep-Live-Cam-Server
cd Deep-Live-Cam-Server

2. Crie e ative um ambiente virtual:

python -m venv venv

# Windows
venv\Scripts\activate

# Linux/MacOS
source venv/bin/activate

3. Instale as dependências:

pip install -r requirements.txt

4. Baixe os modelos necessários:

• GFPGANv1.4
• inswapper_128_fp16.onnx

Coloque os arquivos na pasta "models".

Configuração GPU (NVIDIA)

1. Instale o CUDA Toolkit 11.8.0
2. Configure o onnxruntime-gpu:

pip uninstall onnxruntime onnxruntime-gpu
pip install onnxruntime-gpu==1.16.3

Uso

Servidor

1. Inicie o servidor WebSocket:

python server_ws.py

O servidor iniciará na porta 8765 por padrão.

Cliente

1. Configure o endereço do servidor no arquivo de configuração
2. Execute o cliente:

python client.py

Configuração em Nuvem

Para melhor performance, recomendamos hospedar o servidor em uma máquina com GPU dedicada. Algumas opções:

• Google Cloud com GPUs NVIDIA T4
• AWS EC2 com instâncias g4dn
• Servidores dedicados com GPUs

---

使用腾讯云GPU服务器 + ubuntu 22.04 + 不预装 驱动 使用到的文件 都可以在服务器上 19.143.14.205:8011/xxxxxx 上面wget下载

1. https://blog.csdn.net/xundh/article/details/127974227 直接开始第二步 安装cuda，因为cuda 安装程序 带 驱动。

2. 下载 cuda_11.8.0_520.61.05_linux.run

3. wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run

4. sudo sh cuda_11.8.0_520.61.05_linux.run

5. 勾选上驱动，然后安装

   export PATH=$PATH:/usr/local/cuda-11.8/bin export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.8/lib64 sudo ldconfig

6. 测试 nvcc --version ; nvidia-smi

7. 下载cudnn 8.6 https://developer.nvidia.com/rdp/cudnn-archive, 下载到本地

8. sudo dpkg -i cudnn-local-repo-ubuntu2204-8.6.0.163_1.0-1_amd64.deb sudo cp /var/cudnn-local-repo-ubuntu2204-8.6.0.163/cudnn-local-FAED14DD-keyring.gpg /usr/share/keyrings/ sudo apt-get update sudo apt-get install libcudnn8=8.6.0.163-1+cuda11.8 # 这里输入到=按tab补全即可，安装运行库 sudo apt-get install libcudnn8-dev=8.6.0.163-1+cuda11.8 # 通过tab自动补全，安装developer 库 sudo apt-get install libcudnn8-samples=8.6.0.163-1+cuda11.8 # 通过tab自动补全，安装示例和文档

或者可以离线安装cudnn

CUDNN_TAR_FILE="cudnn-linux-x86_64-8.7.0.84_cuda11-archive.tar.xz" sudo wget https://developer.download.nvidia.com/compute/redist/cudnn/v8.7.0/local_installers/11.8/cudnn-linux-x86_64-8.7.0.84_cuda11-archive.tar.xz sudo tar -xvf ${CUDNN_TAR_FILE} sudo mv cudnn-linux-x86_64-8.7.0.84_cuda11-archive cuda

sudo cp -P cuda/include/cudnn.h /usr/local/cuda-11.8/include sudo cp -P cuda/lib/libcudnn* /usr/local/cuda-11.8/lib64/ sudo chmod a+r /usr/local/cuda-11.8/lib64/libcudnn*

8.测试 cudnn cp -r /usr/src/cudnn_samples_v8/ $HOME cd ~/cudnn_samples_v8/mnistCUDNN/ make clean && make sudo apt-get install libfreeimage3 libfreeimage-dev ./mnistCUDNN

9.下载 minconda https://mirrors.tuna.tsinghua.edu.cn/anaconda/miniconda/Miniconda3-py311_25.7.0-2-Linux-x86_64.sh 10. ./conda init 11. conda create -n py311 python==3.11 12. conda activate py311

13. 安装ffmpeg sudo apt install ffmpeg

14. 准备安装 requirements.txt

15. pip install cython pip install pysocks pip install -i https://mirrors.aliyun.com/pypi/simple tb-nightly pip install --upgrade pip setuptools wheel pip install torch==2.6.0 --index-url https://download.pytorch.org/whl/cu118 pip install -i https://pypi.tuna.tsinghua.edu.cn/simple basicsr==1.4.2

16. 安装 requirements.txt，内容如下 cog==0.14.12 customtkinter==5.2.2 cv2_enumerate_cameras==1.1.18.3 facexlib==0.3.0 insightface==0.7.3 numpy<2 onnxruntime_gpu==1.16.3 opencv_python_headless==4.11.0.86 opennsfw2==0.14.0 Pillow==11.2.1 pygrabber==0.2 PyYAML==6.0.2 realesrgan==0.3.0 scikit_learn==1.2.2 tensorflow==2.20.0 torch==2.7.0 torchvision tqdm==4.66.4 websocket_client==1.8.0 websockets==12.0

    #torch_tensorrt==2.7.0 装不上的按下面命令装 pip install torch torchvision torchaudio torch_tensorrt --index-url https://download.pytorch.org/whl/cu118

    pip install -r requirements.txt --index-url https://download.pytorch.org/whl/cu118

17. pip install -r requirements.txt

18. 启动程序前 第一次 先设置个 代理， 因为要下载些东西 export http_proxy="socks5://8YZsweP:Rtw2111j@18.55.140.95:10120" export https_proxy="socks5://8YZsweP:Rtw2111j@18.55.140.95:10120" export http_proxy=;export https_proxy=

19. 修改 /root/miniconda3/envs/py311/lib/python3.11/site-packages/basicsr/data/degradations.py 第8行 为 from torchvision.transforms._functional_tensor import rgb_to_grayscale

20. 从服务器上下载那两个模型 inswapper_128_fp16.onnx GFPGANv1.4.pth 到 models 文件夹

21. python server_ws.py

22. 客户端 本地台式机安装 pip install websocket_client==1.8.0 pip install websocket_client==1.8.0 pip install opencv_python==4.8.0.74 python client_ws.py

23. 如果遇到 TensorrtExecutionProvider 错误 wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/8.6.1/tars/TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz tar -xzf TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz sudo mv TensorRT-8.6.1.6 /usr/local/tensorrt echo 'export LD_LIBRARY_PATH=/usr/local/tensorrt/lib:$LD_LIBRARY_PATH' >> ~/.bashrc echo 'export PATH=/usr/local/tensorrt/bin:$PATH' >> ~/.bashrc source ~/.bashrc ls /usr/local/tensorrt/lib/libnvinfer.so.8

    ---

    最新版的怎么安装 tensorrt wget https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb dpkg -i cuda-keyring_1.1-1_all.deb apt-get clean apt-get update

    curl -fsSL https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2204/x86_64/3bf863cc.pub | gpg --dearmor > /usr/share/keyrings/nvidia-cuda-archive-keyring.gpg cat >/etc/apt/sources.list.d/cuda.list <<'EOF' deb [signed-by=/usr/share/keyrings/nvidia-cuda-archive-keyring.gpg] https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2204/x86_64/ / EOF apt-get clean apt-get update

    apt-get install -y libnvonnxparsers10

24. 出现ModuleNotFoundError: No module named 'torchvision.transforms.functional_tensor'的原因大概是原先的“名字”改了，但是安装的basicsr包中的名字没有改，所以会报错。 只要在miniconda3/lib/python3.12/site-packages/basicsr/data/degradations.py文件中第8行将 原from torchvision.transforms.functional_tensor import rgb_to_grayscale 改成from torchvision.transforms._functional_tensor import rgb_to_grayscale 或者改成from torchvision.transforms.functional import rgb_to_grayscale 均能够解决问题

25. 5090显卡上 onnx 最新版有问题 ， 需要使用 pip install onnxruntime-gpu==1.22.0 cog==0.14.12 customtkinter==5.2.2 cv2_enumerate_cameras==1.1.18.3 facexlib insightface numpy<2 onnxruntime-gpu==1.22.0 opencv_python_headless==4.11.0.86 opennsfw2==0.14.0 Pillow pygrabber PyYAML==6.0.2 realesrgan==0.3.0 scikit_learn==1.2.2 tensorflow torch torch_tensorrt torchvision websocket_client==1.8.0 websockets==12.0

26. Instructions for CUDA v11.8 and cuDNN 8.7 installation on Ubuntu 22.04 for PyTorch 2.0.0 https://gist.github.com/lamcnguyen89/71ba818f9492d1b7d5f45b1093e78863

---

server_ws.py 的一些优化

1.开启人脸增强的开关

from modules.processors.frame import face_enhancer new_face = face_enhancer.enhance_face(new_face)

2.增大frame 交换 FaceSwapServer(max_workers=30)

3.关闭 TensorrtExecutionProvider 优化加速 #if 'TensorrtExecutionProvider' in providers:

return ['TensorrtExecutionProvider']

4.返回给客户端 png，增强画质(好像不起作用) #_, buffer = cv2.imencode('.jpg', frame) _, buffer = cv2.imencode('.png', frame)