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:
        client = 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), input_bucket_name=Header(None), output_bucket_name=Header(None), write_to_gcs=Header(False), examination_id=Header(None))

POST: /bucket_invocations

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

Headers

Input-Location: Location of the image / folder in the GCS bucket.

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.

Write-To-GCS: Bool flag to write the results to a GCS bucket. False by default.

Examination-ID: ID of the examination, used to track the request results.

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

- scan_width: width of the scan window.

- mm_crop_zone: how much to crop from the center of the image.

- center_coordinates: center coordinates of the image (obtained from fovea center model).

- pixel_spacing_row: pixel spacing row parameter of the exam.

- pixel_spacing_column: pixel spacing column parameter of the exam.

- low_confidence_p: low confidence probability level, initially set to 0.1.

- nerve_zone_landmarks: optional, landmarks of the nerve zone returned by retinal_app

- nerve_zone_slice_indices: optional, slice indices of the nerve zone returned by retinal_app

- mm_crop_zone_nerve: how much to crop from the nerve of the image, initially set to 1.0.

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 or webhook url is missing.

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),
    input_bucket_name: str = Header(None),
    output_bucket_name: str = Header(None),
    write_to_gcs: bool = Header(False),
    examination_id: str = Header(None),
):
    """
    # POST: /bucket_invocations

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

    Headers:
        *Input-Location*: Location of the image / folder in the GCS bucket.

        *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.

        *Write-To-GCS*: Bool flag to write the results to a GCS bucket. False by default.

        *Examination-ID*: ID of the examination, used to track the request results.

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

            - scan_width: width of the scan window.

            - mm_crop_zone: how much to crop from the center of the image.

            - center_coordinates: center coordinates of the image (obtained from fovea center model).

            - pixel_spacing_row: pixel spacing row parameter of the exam.

            - pixel_spacing_column: pixel spacing column parameter of the exam.

            - low_confidence_p: low confidence probability level, initially set to 0.1.

            - nerve_zone_landmarks: optional, landmarks of the nerve zone returned by retinal_app

            - nerve_zone_slice_indices: optional, slice indices of the nerve zone returned by retinal_app

            - mm_crop_zone_nerve: how much to crop from the nerve of the image, initially set to 1.0.

    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 or webhook url is missing.
    """
    webhook_response = _check_webhook(webhook_url, examination_id, 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
    )
    images = _read_from_gcp_bucket(input_bucket_to_use, input_location, examination_id, logger)

    if not images:
        return JSONResponse(content={"error": "No images found", "examination_id": examination_id}, 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["filenames"],
                "request_uuid": response["request_uuids"],
                "result_path": response["result_paths"],
                "status": "sent",
                "examination_id": examination_id,
            },
            status_code=200,
        )
    except Exception as e:
        log_message = json.dumps(
            {
                "status": "WARNING",
                "message": f"Unnexpected error while processing images: {str(e)}",
                "examination_id": examination_id,
            }
        )
        logger.warning(log_message)

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

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

POST: /bucket_invocations_azure_uae

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

Headers

Input-Location: Location of the image / folder in the Azure Blob bucket.

Webhook-Url: URL to send the results of the inference.

Input-Bucket-Name: Name of the input Azure Blob bucket. Default is the bucket name in the config file.

Output-Bucket-Name: Name of the output Azure Blob bucket. Default is the bucket name in the config file.

Write-To-GCS: Bool flag to write the results to a GCS bucket. False by default.

Examination-ID: ID of the examination, used to track the request results.

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

- scan_width: width of the scan window.

- mm_crop_zone: how much to crop from the center of the image.

- center_coordinates: center coordinates of the image (obtained from fovea center model).

- pixel_spacing_column: pixel spacing column parameter of the exam.

- pixel_spacing_row: pixel spacing row parameter of the exam.

- low_confidence_p: low confidence probability level, initially set to 0.1.

- nerve_zone_landmarks: optional, landmarks of the nerve zone returned by retinal_app

- nerve_zone_slice_indices: optional, slice indices of the nerve zone returned by retinal_app

- mm_crop_zone_nerve: how much to crop from the nerve of the image, initially set to 1.0.

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 Azure Blob 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 or webhook url is missing.

Source code in app.py

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

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

    Headers:
        *Input-Location*: Location of the image / folder in the Azure Blob bucket.

        *Webhook-Url*: URL to send the results of the inference.

        *Input-Bucket-Name*: Name of the input Azure Blob bucket. Default is the bucket name in the config file.

        *Output-Bucket-Name*: Name of the output Azure Blob bucket. Default is the bucket name in the config file.

        *Write-To-GCS*: Bool flag to write the results to a GCS bucket. False by default.

        *Examination-ID*: ID of the examination, used to track the request results.

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

            - scan_width: width of the scan window.

            - mm_crop_zone: how much to crop from the center of the image.

            - center_coordinates: center coordinates of the image (obtained from fovea center model).

            - pixel_spacing_column: pixel spacing column parameter of the exam.

            - pixel_spacing_row: pixel spacing row parameter of the exam.

            - low_confidence_p: low confidence probability level, initially set to 0.1.

            - nerve_zone_landmarks: optional, landmarks of the nerve zone returned by retinal_app

            - nerve_zone_slice_indices: optional, slice indices of the nerve zone returned by retinal_app

            - mm_crop_zone_nerve: how much to crop from the nerve of the image, initially set to 1.0.

    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 Azure Blob 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 or webhook url is missing.
    """

    webhook_response = _check_webhook(webhook_url, examination_id, 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.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_uae_bucket(
        input_bucket_to_use, 
        input_location, 
        config.azure_uae_connection_string, 
        examination_id, 
        logger
    )

    if not images:
        return JSONResponse(
            content={"error": "No images found", "examination_id": examination_id}, 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["filenames"],
                "request_uuid": response["request_uuids"],
                "result_path": response["result_paths"],
                "status": "sent",
                "examination_id": examination_id,
            },
            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

Input-Location: Location of the image in the GCS bucket.

Webhook-Url: URL to send the results of the inference.

Examination-ID: ID of the examination, used to track the request results.

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

- scan_width: width of the scan window.

- mm_crop_zone: how much to crop from the center of the image.

- center_coordinates: center coordinates of the image (obtained from fovea center model).

- pixel_spacing_row: pixel spacing row parameter of the exam.

- pixel_spacing_column: pixel spacing column parameter of the exam.

- low_confidence_p: low confidence probability level, initially set to 0.1.

- slice_idx: optional, index of the slice to process, needed for nerve zone cropp.

- nerve_zone_landmarks: optional, landmarks of the nerve zone returned by retinal_app

- nerve_zone_slice_indices: optional, slice indices of the nerve zone returned by retinal_app

- mm_nerve_crop_zone: how much to crop from the nerve of the image, initially set to 1.0.

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 or webhook url is missing.

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:
        *Input-Location*: Location of the image in the GCS bucket.

        *Webhook-Url*: URL to send the results of the inference.

        *Examination-ID*: ID of the examination, used to track the request results.

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

            - scan_width: width of the scan window.

            - mm_crop_zone: how much to crop from the center of the image.

            - center_coordinates: center coordinates of the image (obtained from fovea center model).

            - pixel_spacing_row: pixel spacing row parameter of the exam.

            - pixel_spacing_column: pixel spacing column parameter of the exam.

            - low_confidence_p: low confidence probability level, initially set to 0.1.

            - slice_idx: optional, index of the slice to process, needed for nerve zone cropp.

            - nerve_zone_landmarks: optional, landmarks of the nerve zone returned by retinal_app

            - nerve_zone_slice_indices: optional, slice indices of the nerve zone returned by retinal_app

            - mm_nerve_crop_zone: how much to crop from the nerve of the image, initially set to 1.0.

    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 or webhook url is missing.
    """

    client = grpcclient.InferenceServerClient(
        url=config.grcp_model_server_address,
        verbose=False,
        channel_args=(("grpc.lb_policy_name", "round_robin"),),
    )
    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)
    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[np.newaxis, ..., np.newaxis]

        inputs = [
            grpcclient.InferInput("IMAGE", img.shape, np_to_triton_dtype(img.dtype)),
            grpcclient.InferInput("INPUT_JSON", (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,
                request_uuid=request_uuid,
                filename=image.filename,
                webhook_url=webhook_url,
                examination_id=examination_id,
                client=client,
                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",
                "request_uuid": request_uuid,
                "examination_id": examination_id,
            },
            status_code=400,
        )