Collection of Docker images for ML/DL and video processing projects.

Overview of images

Several types of images are presented:

• core: CUDA (12.2.2), cuDNN (8.9.5), Python (3.11.6)
• base: Python ML and CV packages, FFmpeg (6.0) with NVENC/NVDEC support
• pytorch: TensorRT (8.6.1), PyTorch (2.1.0), torchvision (0.16.0), torchaudio (2.1.0) and torch based libraries
• video: VideoProcessingFramework (2.0.0) and Tensor Stream (0.5.0) for full HW acceleration of video decoding

Supported NVIDIA architectures: Pascal (sm_60, sm_61), Volta (sm_70), Turing (sm_75), Ampere (sm_80, sm_86), Ada Lovelace (sm_89), Hopper (sm_90).
You can identify the architecture of your GPU here.

Example

Pull an image from the GitHub Container registry

docker pull ghcr.io/osai-ai/dokai:23.10-pytorch

Docker Hub mirror

docker pull osaiai/dokai:23.10-pytorch

Check available GPUs inside the container

docker run --rm \
    --gpus=all \
    ghcr.io/osai-ai/dokai:23.10-pytorch \
    nvidia-smi

Package versions

dokai:23.10-core

ghcr.io/osai-ai/dokai:23.10-core

CUDA (12.2.2), cuDNN (8.9.5)
Python (3.11.6) CMake (3.22.1) Bazelisk (1.18.0)

dokai:23.10-base

ghcr.io/osai-ai/dokai:23.10-base

additionally to dokai:23.10-core:

FFmpeg (n6.0), nv-codec-headers (n12.1.14.0)

packaging==23.2
numpy==1.26.1
opencv-python==4.8.1.78
sympy==1.12
scipy==1.11.3
matplotlib==3.8.0
pandas==2.1.1
scikit-learn==1.3.2
scikit-image==0.22.0
Pillow==10.1.0
av==10.0.0
librosa==0.10.1
albumentations==1.3.1
pyzmq==25.1.1
Cython==3.0.4
numba==0.58.1
requests==2.31.0
psutil==5.9.6
pydantic==2.4.2
PyYAML==6.0.1
notebook==7.0.6
ipywidgets==8.1.1
tqdm==4.66.1
pytest==7.4.2
pytest-cov==4.1.0
mypy==1.6.1
flake8==6.1.0
pre-commit==3.5.0

dokai:23.10-pytorch

ghcr.io/osai-ai/dokai:23.10-pytorch

additionally to dokai:23.10-base:

TensorRT (8.6.1)
MAGMA (2.7.1)

torch==2.1.0 (source, v2.1.0)
torch_tensorrt (source, v2.1.0-rc3)
torchvision==0.16.0 (source, v0.16.0)
torchaudio==2.1.0 (source, v2.1.0)
triton==2.1.0
cupy-cuda12x==12.2.0
pytorch-ignite==0.4.13
pytorch-argus==1.0.0
kornia==0.7.0
timm==0.9.8
onnx==1.14.1
onnxruntime==1.16.1
onnxsim==0.4.35

dokai:23.10-video

ghcr.io/osai-ai/dokai:23.10-video

additionally to dokai:23.10-pytorch:

VideoProcessingFramework==2.0.0 (source, v2.0.0)
tensor-stream==0.5.0 (source, 0.5.0)

Versions for past releases can be found in release descriptions.