Início / user-2549330

Emmanuel David A's questions

Martin Hope
Emmanuel David A
Asked: 2025-01-30 15:54:37 +0800 CST
  • 6

Estou tentando executar ffmpeg com suporte cuda no docker. Este é meu Dockerfile

FROM nvidia/cuda:12.2.0-devel-ubuntu22.04

ENV DEBIAN_FRONTEND=noninteractive
RUN ln -fs /usr/share/zoneinfo/UTC /etc/localtime && \
    echo "UTC" > /etc/timezone && \
    apt-get update -y && apt-get install -y tzdata && \
    dpkg-reconfigure -f noninteractive tzdata

# Install system dependencies and build tools
RUN apt-get update -qq && apt-get -y install \
    gcc \
    make \
    autoconf \
    automake \
    build-essential \
    cmake \
    git-core \
    libass-dev \
    libfreetype6-dev \
    libharfbuzz-dev \
    libgnutls28-dev \
    libmp3lame-dev \
    libsdl2-dev \
    libtool \
    libva-dev \
    libvdpau-dev \
    libvorbis-dev \
    libxcb1-dev \
    libxcb-shm0-dev \
    libxcb-xfixes0-dev \
    meson \
    ninja-build \
    pkg-config \
    texinfo \
    wget \
    yasm \
    zlib1g-dev \
    libtool \
    libc6 \
    libc6-dev \
    unzip \
    wget \
    libnuma1 \
    libnuma-dev \
    unzip \
    awscli \
    curl \
    nasm \
    nvidia-cuda-toolkit \
    libnvidia-decode-525-server && \
    apt-get clean

RUN curl -fsSL https://deb.nodesource.com/setup_20.x | bash - && \
    apt-get install -y nodejs && \
    node -v && npm -v

ENV PATH="/usr/local/cuda/bin:${PATH}"
ENV LD_LIBRARY_PATH="/usr/local/cuda/lib64:${LD_LIBRARY_PATH}"

# Install nv-codec-headers
RUN mkdir -p ~/ffmpeg_sources ~/bin && \
    cd ~/ffmpeg_sources && \
    git clone https://git.videolan.org/git/ffmpeg/nv-codec-headers.git && \
    cd nv-codec-headers && \
    make install

# Install FFmpeg with CUDA support
RUN mkdir -p ~/ffmpeg_sources ~/bin && \
    cd ~/ffmpeg_sources && \
    wget -O ffmpeg-snapshot.tar.bz2 https://ffmpeg.org/releases/ffmpeg-snapshot.tar.bz2 && \
    tar xjvf ffmpeg-snapshot.tar.bz2 && \
    cd ffmpeg && \
    PATH="$HOME/bin:$PATH" PKG_CONFIG_PATH="$HOME/ffmpeg_build/lib/pkgconfig" ./configure \
        --enable-nonfree \
        --enable-cuda \
        --enable-cuvid \
        --enable-nvdec \
        --enable-nvenc \
        --enable-cuda-nvcc \
        --enable-libnpp \
        --enable-libfreetype \
        --enable-libharfbuzz \
        --enable-libass \
        --extra-cflags=-I/usr/local/cuda/include \
        --extra-ldflags=-L/usr/local/cuda/lib64 \
        --disable-static \
        --enable-shared && \
    PATH="$HOME/bin:$PATH" make -j$(nproc) && \
    make install && \
    hash -r

RUN cd ~ && mkdir input_videos && mkdir output_videos && mkdir assets

ENV CUDA_VISIBLE_DEVICES=0

CMD ["bash"]

Aqui está meu docker-compose.yml

services:
  ffmpeg-cuda:
    image: ffmpeg-cuda-node
    runtime: nvidia
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]
    volumes:
      - /root/input_videos/:/root/input_videos/
      - /root/output_videos/:/root/output_videos/
      - /root/assets/:/root/assets/

Sempre que executo um comando ffmpeg com cuda recebo este erro

[mjpeg_cuvid @ 0x5df9a07fd880] ctx->cvdl->cuvidGetDecoderCaps(&ctx->caps8) failed -> CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
[mjpeg_cuvid @ 0x5df9a07fd880] ctx->cvdl->cuvidGetDecoderCaps(&ctx->caps10) failed -> CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
[mjpeg_cuvid @ 0x5df9a07fd880] ctx->cvdl->cuvidGetDecoderCaps(&ctx->caps12) failed -> CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

Mas os drivers foram instalados corretamente. nvidia-smi funciona corretamente dentro do contêiner. Por favor, ajude-me a encontrar o que estou esquecendo nisso.

linux
Martin Hope
Emmanuel David A
Asked: 2025-01-24 15:07:32 +0800 CST
  • 6

Eu coloco marca d'água em vídeos como este:

ffmpeg -y -vsync 0 -hwaccel cuda -hwaccel_output_format cuda -i ./input.avi -i ./watermark.png -filter_complex "
[0]hwdownload,format=nv12[downloaded_video];
[1]format=rgba,colorchannelmixer=aa=0.3[logo];
[logo]scale=300:300[scaled_logo];
[downloaded_video][scaled_logo]overlay=W-w-10:H-h+10[overlayed_video];
[overlayed_video]hwupload_cuda=extra_hw_frames=5" -c:v h264_nvenc ./output.mp4

Existe alguma maneira de usar algum dos filtros de escala da nvidia (scale_cuda, scale_npp, scale2ref_npp) e filtro de sobreposição (overlay_cuda) para fazer isso evitando hwdownload e hwupload? Tentei de várias maneiras e não consegui preservar a transparência da imagem png, o que torna a marca d'água um grande bloco branco.

ffmpeg