Darknet is an open source neural network framework written in C. It is fast, efficient, and supports CPU and GPU computation.
I am using NVIDIA Quadro RTX 6000/8000 GPU. This is the first time, I am using darknet framework to develop deep learning model. Here, are the steps to follow to setup darknet framework.
darknet source code : https://github.com/pjreddie/darknet
https://github.com/AlexeyAB/darknet
https://github.com/arnoldfychen/darknet
Installation guide : https://pjreddie.com/darknet/install/
https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html
Supporting resources : pjreddie/darknet#1347
pjreddie/darknet#1874
GPU : NVIDIA Quadro RTX 6000/8000
CUDA toolkit version : 11.4
Driver version : 470.57.02
cuDNN version : 8.8.1.3
OpenCV version : 4.2.0
Ubuntu : 20.04
- Create virtual environment
- Replace myenv with your environment name : virtualenv myenv
- Activate your virtual environment : source myenv/bin/activate
- Install supporting packages
- Install OpenCV library to handle Image IO tasks : sudo apt install libopencv-dev
- Check version by 'pkg-config --modversion opencv4' command
- Install CUDA toolkit according to your GPU
- Run 'nvidia-smi' command to check your CUDA version
- Download cuDNN package (linux tar.xz) according to CUDA version and install it. Documentation is given by cuDNN Documentation (Nvidia).
- Verify installation using mnistCUDNN sample
- Install darknet
- Download darknet using : git clone https://github.com/arnoldfychen/darknet.git
- cd darknet
- Replace the src/image_opencv.cpp file with given file (this will solve compatibility issues of darknet framework with OpenCV 4.0 package)
- Replace Makefile with the given Makefile. Change NVCC=/usr/local/cuda/bin/nvcc by your nvcc path (this will solve error of unsupported gpu architecture)
- At the end test darknet, by running ./darknet imtest data/eagle.jpg