Home

ping()

GET: /ping

Endpoint to check if the server is running.

Returns:

Name	Type	Description
Response		Response with status 200 if the server is running.

Source code in app.py

@app.get("/ping")
def ping():
    """
    # GET: /ping

    Endpoint to check if the server is running.

    Returns:
        Response: Response with status 200 if the server is running.
    """
    try:
        _ = grpcclient.InferenceServerClient(url=config.grcp_model_server_address, verbose=False)
        return Response(status_code=200)
    except Exception:
        return Response(status_code=400)

predict_bucket(input_location=Header(None), inference_parameters=Header(None), webhook_url=Header(None), write_to_gcs=Header(False), input_bucket_name=Header(None), output_bucket_name=Header(None), examination_id=Header(None))

POST: /bucket_invocations

Endpoint to process an image and send it to the inference server.

Headers

Input-Location: Location of the image in the GCS bucket. Webhook-Url: URL to send the results of the inference. Write-To-GCS: Bool flag to write the results to a GCS bucket. False by default. Input-Bucket-Name: Name of the input bucket. Output-Bucket-Name: Name of the output bucket. Examination-ID: ID of the examination, used for logging and tracking. Inference-Parameters: Parameters to send to the inference server. JSON string with the following keys

- threshold: Threshold to apply to the output of the model. Default is 0.25.

- slice_idx: list of slice indices to run, e.g [45, 46, 47, ...]

Returns:

Name	Type	Description
		JSON with the results of the inference:
filename		Name of the file that was processed. >1 if multiple files.
status		Status of the request. Can be "sent" or "error".
result_path		Path to the result in the GCS bucket. >1 if multiple files.
request_uuid		UUID of the request, generated by the server. Used to track the request results. >1 if multiple files, in correspondence with the filename.

Raises:

Type	Description
Response	Error response if the content type is not supported.

Source code in app.py

@app.post("/bucket_invocations")
def predict_bucket(
    input_location: str = Header(None),
    inference_parameters: str = Header(None),
    webhook_url: str = Header(None),
    write_to_gcs: bool = Header(False),
    input_bucket_name: str = Header(None),
    output_bucket_name: str = Header(None),
    examination_id: str = Header(None),
):
    """
    # POST: /bucket_invocations

    Endpoint to process an image and send it to the inference server.

    Headers:
        *Input-Location*: Location of the image in the GCS bucket.
        *Webhook-Url*: URL to send the results of the inference.
        *Write-To-GCS*: Bool flag to write the results to a GCS bucket. False by default.
        *Input-Bucket-Name*: Name of the input bucket.
        *Output-Bucket-Name*: Name of the output bucket.
        *Examination-ID*: ID of the examination, used for logging and tracking.
        *Inference-Parameters*: Parameters to send to the inference server. JSON string with the following keys

            - threshold: Threshold to apply to the output of the model. Default is 0.25.

            - slice_idx: list of slice indices to run, e.g [45, 46, 47, ...]

    Returns:
        JSON with the results of the inference:
        filename: Name of the file that was processed. >1 if multiple files.
        status: Status of the request. Can be "sent" or "error".
        result_path: Path to the result in the GCS bucket. >1 if multiple files.
        request_uuid: UUID of the request, generated by the server. Used to track the request results. >1 if multiple files, in correspondence with the filename.

    Raises:
        Response: Error response if the content type is not supported.
    """

    webhook_response = _check_webhook(webhook_url, examination_id=examination_id, logger=logger)
    if webhook_response.status_code != 200:
        return webhook_response

    # get input and output locations
    input_bucket_to_use = (
        input_bucket_name if input_bucket_name is not None else config.input_bucket_name
    )
    output_bucket_to_use = (
        output_bucket_name if output_bucket_name is not None else config.output_bucket_name
    )
    if inference_parameters is None:
        inference_parameters = "{'threshold': 0.25}"
    inference_parameters = inference_parameters.replace("'", '"')
    inference_params = json.loads(inference_parameters)

    slice_idx = None
    if "slice_idx" in inference_params.keys():
        slice_idx = inference_params["slice_idx"]
        # add a lead 0
        slice_idx = [f"{i:03d}" for i in slice_idx]
        del inference_params["slice_idx"]

    start = time.time()
    images = asyncio.run(
        _read_from_gcp_bucket_async(input_bucket_to_use, input_location, examination_id, config, slice_idx=slice_idx, logger=logger)
    )
    elapsed = time.time() - start

    logger.info(json.dumps({
        "status": "INFO",
        "message": f"Read {len(images)} images from GCP bucket {input_bucket_to_use}/{input_location}. Took {elapsed} seconds.",
        "examination_id": examination_id
    }))


    try:
        response = _process_images(
            images,
            inference_parameters,
            output_bucket_to_use,
            webhook_url,
            config,
            write_to_gcs,
            examination_id,
            logger=logger
        )

        return JSONResponse(
            content={
                "filename": response["filename"],
                "request_uuid": response["request_uuid"],
                "result_path": response["result_path"],
                "examination_id": examination_id,
                "status": "sent",
            },
            status_code=200,
        )

    except Exception as e:
        return JSONResponse(
            content={"error": str(e), "examination_id": examination_id}, status_code=400
        )

predict_bucket_azure_uae(input_location=Header(None), webhook_url=Header(None), inference_parameters=Header(None), write_to_gcs=Header(False), input_bucket_name=Header(None), output_bucket_name=Header(None), examination_id=Header(None))

POST: /bucket_invocations_azure_uae

Endpoint to process images from Microsoft Azure Blob Storage (UAE).

Headers

Input-Location: Location of the image in the bucket. Examination-ID: ID of the examination, used for logging and tracking. Webhook-Url: URL to send the results of the inference. Input-Bucket-Name: Name of the input GCS bucket. Default is the bucket name in the config file. Output-Bucket-Name: Name of the output GCS bucket. Default is the bucket name in the config file. Inference-Parameters: Parameters to send to the inference server. JSON string with the following keys

- threshold: Threshold to apply to the output of the model. Default is 0.25.

Returns:

Name	Type	Description
		JSON with the results of the inference:
filename		Name of the file that was processed. >1 if multiple files.
status		Status of the request. Can be "sent" or "error".
result_path		Path to the result in the GCS bucket. >1 if multiple files.
request_uuid		UUID of the request, generated by the server. Used to track the request results. >1 if multiple files, in correspondence with the filename.

Raises:

Type	Description
Response	Error response if the content type is not supported.

Source code in app.py

@app.post("/bucket_invocations_azure_uae")
def predict_bucket_azure_uae(
    input_location: str = Header(None),
    webhook_url: str = Header(None),
    inference_parameters: str = Header(None),
    write_to_gcs: bool = Header(False),
    input_bucket_name: str = Header(None),
    output_bucket_name: str = Header(None),
    examination_id: str = Header(None),
):
    """
    # POST: /bucket_invocations_azure_uae

    Endpoint to process images from Microsoft Azure Blob Storage (UAE).

    Headers:
        *Input-Location*: Location of the image in the bucket.
        *Examination-ID*: ID of the examination, used for logging and tracking.
        *Webhook-Url*: URL to send the results of the inference.
        *Input-Bucket-Name*: Name of the input GCS bucket. Default is the bucket name in the config file.
        *Output-Bucket-Name*: Name of the output GCS bucket. Default is the bucket name in the config file.
        *Inference-Parameters*: Parameters to send to the inference server. JSON string with the following keys


            - threshold: Threshold to apply to the output of the model. Default is 0.25.

    Returns:
        JSON with the results of the inference:
        filename: Name of the file that was processed. >1 if multiple files.
        status: Status of the request. Can be "sent" or "error".
        result_path: Path to the result in the GCS bucket. >1 if multiple files.
        request_uuid: UUID of the request, generated by the server. Used to track the request results. >1 if multiple files, in correspondence with the filename.

    Raises:
        Response: Error response if the content type is not supported.
    """
    webhook_response = _check_webhook(webhook_url, examination_id=examination_id, logger=logger)
    if webhook_response.status_code != 200:
        return webhook_response

    input_bucket_to_use = (
        input_bucket_name if input_bucket_name is not None else config.azure_uae_input_bucket_name
    )
    output_bucket_to_use = (
        output_bucket_name
        if output_bucket_name is not None
        else config.azure_uae_output_bucket_name
    )

    images = _read_from_azure_blob(
        input_bucket_to_use,
        input_location,
    )

    if not images:
        return JSONResponse(content={"error": "No images found"}, status_code=400)

    try:
        response = _process_images(
            images,
            inference_parameters,
            output_bucket_to_use,
            webhook_url,
            config,
            write_to_gcs,
            examination_id,
            logger=logger
        )

        return JSONResponse(
            content={
                "filename": response["filename"],
                "request_uuid": response["request_uuid"],
                "result_path": response["result_path"],
                "examination_id": examination_id,
                "status": "sent",
            },
            status_code=200,
        )

    except Exception as e:
        return JSONResponse(
            content={"error": str(e), "examination_id": examination_id}, status_code=400
        )

predict_image(image=File(...), inference_parameters=Header(None), webhook_url=Header(None), examination_id=Header(None))

POST: /invocations

Endpoint to process an image and send it to the inference server.

Parameters:

Name	Type	Description	Default
image	UploadFile	Image file to process (in the request body).	File(...)

Headers

Inference-Parameters: Parameters to send to the inference server. JSON string with the following keys:

- threshold: Threshold to apply to the output of the model. Default is 0.25.

Content-Type: Type of the image. Can be "image/jpeg", "image/png", "image/tiff", "image/bmp", "image/jpg". Webhook-URL: URL to send the results of the inference. Examination-ID: ID of the examination, used for logging and tracking.

Returns:

Type	Description
	JSON with the results of the inference:
	filename: Name of the file that was processed.
	status: Status of the request. Can be "sent" or "error".
	request_uuid: UUID of the request, generated by the server. Used to track the request results.

Raises:

Type	Description
Response	Error response if the content type is not supported.

Source code in app.py

@app.post("/invocations")
def predict_image(
    image: UploadFile = File(...),
    inference_parameters: str = Header(None),
    webhook_url: str = Header(None),
    examination_id: str = Header(None),
):
    """
    # POST: /invocations

    Endpoint to process an image and send it to the inference server.

    Args:
        image (UploadFile): Image file to process (in the request body).

    Headers:
        *Inference-Parameters*: Parameters to send to the inference server. JSON string with the following keys:

            - threshold: Threshold to apply to the output of the model. Default is 0.25.

        *Content-Type*: Type of the image. Can be "image/jpeg", "image/png", "image/tiff", "image/bmp", "image/jpg".
        *Webhook-URL*: URL to send the results of the inference.
        *Examination-ID*: ID of the examination, used for logging and tracking.

    Returns:
        JSON with the results of the inference:
        - filename: Name of the file that was processed.
        - status: Status of the request. Can be "sent" or "error".
        - request_uuid: UUID of the request, generated by the server. Used to track the request results.

    Raises:
        Response: Error response if the content type is not supported.
    """

    client = grpcclient.InferenceServerClient(
        url=config.grcp_model_server_address,
        verbose=False,
        channel_args=(("grpc.lb_policy_name", "round_robin"),),
    )  # , concurrency=1, connection_timeout=10)
    model_config = client.get_model_config(
        model_name=config.model_name, model_version=config.model_version, as_json=True
    )["config"]

    content_type = image.content_type

    webhook_response = _check_webhook(webhook_url, examination_id, logger=logger)
    if webhook_response.status_code != 200:
        return webhook_response

    if content_type not in config.available_content_types:
        return Response(
            status=415,
            content="Cannot decode image data. Is content_type correct?",
            media_type="text/plain",
        )

    try:
        contents = image.file.read()

        image_bytes = np.frombuffer(contents, dtype=np.uint8)

        img = cv2.imdecode(image_bytes, cv2.IMREAD_GRAYSCALE)

        img = img.astype(np.uint8)
        img = img[np.newaxis, ...]

        inputs = [
            grpcclient.InferInput("IMAGE", img.shape, np_to_triton_dtype(img.dtype)),
            grpcclient.InferInput("INPUT_JSON_PARAMS", (1, 1), "BYTES"),
        ]
        inputs[0].set_data_from_numpy(img)
        inference_params = inference_parameters.replace("'", '"')
        inputs[1].set_data_from_numpy(np.array([[inference_params]] * 1, dtype=np.object_))

        outputs = [
            grpcclient.InferRequestedOutput(model_config["output"][i]["name"])
            for i in range(len(model_config["output"]))
        ]

        request_uuid = str(uuid.uuid4())

        _ = client.async_infer(
            model_name=config.model_name,
            model_version=config.model_version,
            inputs=inputs,
            outputs=outputs,
            callback=partial(
                result_callback,
                model_config=model_config,
                filename=image.filename,
                request_uuid=request_uuid,
                client=client,
                webhook_url=webhook_url,
                image_resolution=(img.shape[1], img.shape[2]),
                examination_id=examination_id,
                logger=logger
            ),
        )

        return JSONResponse(
            content={
                "filename": image.filename,
                "status": "sent",
                "request_uuid": request_uuid,
                "examination_id": examination_id,
            },
            status_code=200,
        )
    except Exception as e:
        return JSONResponse(
            content={"message": str(e), "status": "error", "examination_id": examination_id},
            status_code=400,
        )