yolov9-rest-service-xray-fracture

Production ready YoloV9 REST Service for x-ray fracture detection

Running locally

Clone this repo

git clone https://github.com/AzyriAI/yolov9-docker-rest-service-xray-fracture.git
cd yolov9-docker-rest-service-xray-fracture

Install requirements

pip install -r requirements.txt

Download pre-trained weights

In case you didn't train your own model you can download this train. Weights are from best model after 100 epochs.

It tooks me 12 hours to train yolov9 for frature detection (1 GPU)

https://www.kaggle.com/datasets/karelbecerra/yolov9-weights-x-ray-fracture-detection

unzip best.pt.zip
mv best.pt ./weights 

Set up environment variables

You can find enviroment variables in env.sample

export MODEL_STORAGE=./weights/
export SERVER_HOST=http://localhost:3100
export OUT_PUT=./prediction/
export SERVICE=/v1/fracture/prediction/

Run your server

uvicorn src.main:server --host 0.0.0.0 --port 3100

You shoud see a similar output

Fusing layers... 
yolov9-c summary: 962 layers, 51018070 parameters, 0 gradients, 239.0 GFLOPs
INFO:     Azyri AI Server - starting
INFO:     AI Server - init api v1
INFO:     Started server process [61457]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:3100 (Press CTRL+C to quit)

Test your prediction

Open a new terminal. Server expects url as input and returns local url as output.

Example with fra1.png:

curl -X POST http://localhost:3100/v1/fracture/predict \
-H "Content-Type: application/json" \
-d '{"imageUrl": "https://github.com/AzyriAI/ai-ml-dl-samples/blob/main/data/fracture/fra1.png?raw=true"}'

Example with fra2.png:

curl -X POST http://localhost:3100/v1/fracture/predict \
-H "Content-Type: application/json" \
-d '{"imageUrl": "https://github.com/AzyriAI/ai-ml-dl-samples/blob/main/data/fracture/fra2.png?raw=true"}'

You can find some x-rays to test your own local server in https://github.com/AzyriAI/ai-ml-dl-samples/tree/main/data/fracture

Check your result

After running your prediction and if everything went well you should get a response similar to this

{"status":"success","payload":"http://localhost:3100/v1/fracture/prediction/edf66m4r0tj2rsjmknou.png"}

You have to options 1- Open your browser and paste the result url 2- Execute wget command

wget http://localhost:3100/v1/fracture/prediction/edf66m4r0tj2rsjmknou.png

References

Yolov9 https://github.com/WongKinYiu/yolov9