Docker images for Darknet Yolo v4, v3 and v2. AlexeyAB/darknet
All of Darknet's images are based on Ubuntu 20.04. gpu-
images are based on 11.2.2-cudnn8-ubuntu 20.04
Tags means options in the Makefile when darknet binary was built. You can check options here.
cpu
tag means images are built withAVX=1
andOPENMP=1
.noopt
tag means disabling AVX optionAVX=0
. If you use cpu based image and get error, try this tag.
cv
tag means images are built withOPENCV=1
gpu
tag means images are built withGPU=1
andCUDNN=1
cc**
tag means compute compability of GPU. Images with this tag are optimized for GPU architecture. You can check compute compability of your GPU here. If compute compability is greater than or equal to 7.0, images are built withCUDNN_HALF=1
.
This repository build docker images from latest darknet commit automatically. If you want to use released darknet images, please add released tag name before base image tags. For example when you want to use YOLOv4 pre-release gpu image, you can pull image as follows.
$ docker pull daisukekobayashi/darknet:darknet_yolo_v4_pre-gpu
You can check available tags here.
$ docker run --rm -v $PWD:/workspace -w /workspace \
daisukekobayashi/darknet:cpu darknet ...
Start a Darknet cpu container.
$ docker run --runtime=nvidia --rm -v $PWD:/workspace -w /workspace \
daisukekobayashi/darknet:gpu darknet ...
You need to specify nvidia runtime when you use Darknet gpu container.
$ git clone https://github.com/AlexeyAB/darknet.git
$ cd darknet/build/darknet/x64
$ curl -L -O https://pjreddie.com/media/files/yolov3.weights
$ docker run --runtime=nvidia --rm -v $PWD:/workspace -w /workspace daisukekobayashi/darknet:gpu \
darknet detector test data/coco.data yolov3.cfg yolov3.weights -i 0 -thresh 0.25 dog.jpg -ext_output
This is example to test yolov3 using Darknet gpu container.
$ docker run --rm -it daisukekobayashi/darknet:cpu bash -c 'echo "$SOURCE_BRANCH" && echo "$SOURCE_COMMIT"'
Check darknet version.