- A containerized darknet framework with CUDA support. AlexeyAB's fork of darknet is used.
- Instructions for training and inference with YOLOv4 object recognition network.
- Instructions for configuring YOLOv4 for training on custom datasets WIP.
- The weights for the example network which detects dragon heads in images can be found from here.
-
Install Docker
-
Configure darknet to compile without cuda support, unless the PC on which the container is running has the NVIDIA container runtime installed:
Change
use_cuda=1
touse_cuda=0
inDockerfile
-
Download the pre-trained weights and place them into the folder
container/weights/
-
Build and start the docker container:
docker-compose up -d
-
Access the jupyterlab environment from http://localhost:8888/
-
After done, spin the container down:
docker-compose down