Installation¶
Prerequisites¶
eumovie requires several components that must be installed manually before
installing the package itself, because they depend on your specific CUDA version
and require custom builds.
1. NVIDIA GPU and CUDA toolkit¶
A CUDA-capable NVIDIA GPU is required. CUDA 12 or later is recommended. Install the CUDA toolkit from developer.nvidia.com/cuda-downloads.
Verify your installation:
2. CuPy¶
CuPy provides GPU-accelerated array operations. Install the version matching your CUDA toolkit:
See docs.cupy.dev for other versions.
3. OpenCV with CUDA support¶
The standard opencv-python package from PyPI does not include CUDA support.
You must build OpenCV from source with -DWITH_CUDA=ON.
A minimal build sequence (adjust paths for your system):
git clone https://github.com/opencv/opencv.git
cd opencv
mkdir build && cd build
cmake .. \
-DWITH_CUDA=ON \
-DCUDA_FAST_MATH=ON \
-DBUILD_opencv_python3=ON \
-DOPENCV_PYTHON3_INSTALL_PATH=$(python3 -c "import site; print(site.getsitepackages()[0])") \
-DCMAKE_BUILD_TYPE=Release
make -j$(nproc)
sudo make install
Verify CUDA support is available:
4. ffmpeg with NVENC support¶
Install ffmpeg from your system package manager. On Ubuntu:
Verify hardware encoding is available:
Installing eumovie¶
Once the prerequisites are in place:
This installs eumovie as a command-line tool and pulls in the remaining
Python dependencies (numpy, scipy, tifffile) automatically.
Setting up keyframes¶
On first use, generate the keyframes template:
This creates keyframes.py in the current directory and in
~/.config/eumovie/keyframes.py. Edit either file to add camera paths
for your tiles. See the Keyframes documentation for details.
Verifying the installation¶
Should print the full list of options without errors.