An API for Darknet image detection neural networks like YOLO.
The easiest way to use Darkapi is to run the pre-built container:
docker run --rm --name darkapi -p 8080:8080 --device=/dev/nvidiactl --device=/dev/nvidia-uvm --device=/dev/nvidia0 --volume=/opt/nvidia:/usr/local/nvidia:ro squat/darkapiYou can then make requests against the container service, e.g.:
curl 127.0.0.1:8080/api/yolo -F '=@./vendor/darknet/data/dog.jpg'To deploy Darkapi to a Kubernetes cluster, you must have nodes with GPUs and with device plugins installed. Once ready, create the example deployment:
kubectl apply -f kubernetes/deployment.yamlIf you prefer to build Darkapi yourself, first download the sources:
git clone https://github.com/squat/darkapi
cd darkapiBy default, Darkapi is built inside of a NVIDIA Docker container to ensure that the necessary CUDA libraries are present and to guarantee consistent builds.
makeNote: if you have all the necessary CUDA libraries installed locally and wish to run the builds directly on your machine, set CONTAINERIZE=0:
make CONTAINERIZE=0If you want to build Darkapi without GPU support, set GPU=0:
make GPU=0Finally, you will need to download the model weights to be able to use the networks:
make weightsBy default, the Darkapi server runs on port 8080, though this can be configured with the -p flag.
darkapi -p 1337Run the YOLO object detector on an uploaded image. This endpoint expects an image in a multipart upload. Example:
curl 127.0.0.1:8080/api/yolo -F '=@./vendor/darknet/data/dog.jpg'
# {
# "size": 163759,
# "x": 768,
# "y": 576,
# "c": 3,
# "time": 10.835417,
# "detections": [
# {
# "label": "dog",
# "p": 0.990042,
# "x": 0.288539,
# "y": 0.660512,
# "w": 0.243191,
# "h": 0.542467
# },
# {
# "label": "truck",
# "p": 0.92372,
# "x": 0.756574,
# "y": 0.222694,
# "w": 0.280832,
# "h": 0.147699
# },
# {
# "label": "car",
# "p": 0.259004,
# "x": 0.774838,
# "y": 0.224054,
# "w": 0.246296,
# "h": 0.142042
# },
# {
# "label": "bicycle",
# "p": 0.994177,
# "x": 0.473209,
# "y": 0.483861,
# "w": 0.516853,
# "h": 0.576053
# }
# ]
# }Run the YOLO9000 object detector on an uploaded image. This endpoint expects an image in a multipart upload. Example:
curl 127.0.0.1:8080/api/yolo9000 -F '=@./vendor/darknet/data/dog.jpg'
# {
# "size": 163759,
# "x": 768,
# "y": 576,
# "c": 3,
# "time": 7.798958,
# "detections": [
# {
# "label": "Shetland sheepdog",
# "p": 0.563054,
# "x": 0.305005,
# "y": 0.647629,
# "w": 0.23873,
# "h": 0.528642
# },
# {
# "label": "tom",
# "p": 0.402929,
# "x": 0.278132,
# "y": 0.657537,
# "w": 0.296026,
# "h": 0.600402
# },
# {
# "label": "tortoiseshell",
# "p": 0.475844,
# "x": 0.287831,
# "y": 0.654079,
# "w": 0.257309,
# "h": 0.552583
# },
# {
# "label": "push-bike",
# "p": 0.257703,
# "x": 0.514435,
# "y": 0.503511,
# "w": 0.421874,
# "h": 0.384144
# },
# {
# "label": "bicycle-built-for-two",
# "p": 0.570866,
# "x": 0.507706,
# "y": 0.495363,
# "w": 0.455473,
# "h": 0.511342
# },
# {
# "label": "limousine",
# "p": 0.701897,
# "x": 0.731044,
# "y": 0.210221,
# "w": 0.318366,
# "h": 0.172348
# }
# ]
# }Run the tiny YOLO object detector on an uploaded image. This endpoint expects an image in a multipart upload. Example:
curl 127.0.0.1:8080/api/tiny -F '=@./vendor/darknet/data/dog.jpg'
# {
# "size": 163759,
# "x": 768,
# "y": 576,
# "c": 3,
# "time": 1.051016,
# "detections": [
# {
# "label": "dog",
# "p": 0.819004,
# "x": 0.284219,
# "y": 0.689174,
# "w": 0.28744,
# "h": 0.516138
# },
# {
# "label": "car",
# "p": 0.738448,
# "x": 0.74617,
# "y": 0.221734,
# "w": 0.267723,
# "h": 0.161173
# },
# {
# "label": "car",
# "p": 0.380406,
# "x": 0.113937,
# "y": 0.176513,
# "w": 0.04088,
# "h": 0.049404
# },
# {
# "label": "car",
# "p": 0.304245,
# "x": 0.67088,
# "y": 0.208459,
# "w": 0.152685,
# "h": 0.132376
# },
# {
# "label": "bicycle",
# "p": 0.589162,
# "x": 0.423873,
# "y": 0.506519,
# "w": 0.685571,
# "h": 0.512563
# },
# {
# "label": "person",
# "p": 0.267455,
# "x": 0.96026,
# "y": 0.403358,
# "w": 0.07195,
# "h": 0.910909
# },
# {
# "label": "person",
# "p": 0.261761,
# "x": 0.57678,
# "y": 0.176673,
# "w": 0.03534,
# "h": 0.058247
# }
# ]
# }The /healthz endpoint returns a 200 if the API is running.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent