Azure Function + FastAPI + ONNX

main.py

# __init__.py

# import logging
# import azure.functions as func
# from main.app import app


# def main(req: func.HttpRequest, context: func.Context) -> func.HttpResponse:
#     logging.info("Python HTTP trigger function processed a request.")
#     return func.AsgiMiddleware(app).handle(req, context)


import os
import logging
import nest_asyncio
import onnxruntime
from fastapi import FastAPI, UploadFile

max_size = 1152
root = os.path.dirname(os.path.abspath(__file__))
model_path = os.path.join(root, "GL_VGG19_QNRF.onnx")

nest_asyncio.apply()

app = FastAPI()

@app.get("/health")
async def health():
    return {"info": "Up and running!"}

@app.post("/count")
async def count(file: UploadFile):
    logging.info("Converting to PIL...")
    img = Image.open(io.BytesIO(await file.read())).convert("RGB")
    img = ImageOps.exif_transpose(img)
    logging.info("Image size: "+str(img.size))

    if max_size is not None and max(img.size) > max_size:
        logging.info("Resizing image...")
        img.thumbnail((max_size, max_size))

    logging.info("Converting to Numpy...")
    img = np.array(img, dtype=np.float32) / 255.0

    logging.info("Normalizing image...")
    img -= np.array([[[0.485, 0.456, 0.406]]])
    img /= np.array([[[0.229, 0.224, 0.225]]])

    logging.info("Arranging to BCHW...")
    img = img.transpose((2, 0, 1))[None]

    logging.info("Running inference...")
    rt = onnxruntime.InferenceSession(model_path, None)
    mask = rt.run(None, {'input': img})[0][0,0]
    count = int(round(mask.sum()))

    logging.info("Counted {} objects.".format(count))
    return {"count": count}