egslava/test_banana_image_classifier

A PyTorch (image-net) image classifier hosted on banana.dev

ImageNet classifier on Banana.dev

Cloning

Clone the repo and run:

git clone --recurse-submodules https://github.
com/egslava/test_banana_image_classifier.git 

Testing

After cloning, run:

bash run_n_test_all.bash

It will:

1. Create the model from the PyTorch model and the weights
2. Convert it to onnx-format and the resulting high-level interface
3. Test the high-level interface against provided examples
4. Run the API (Banana) server
5. Test the server against the provided examples
6. Build the Docker image
7. Test the Docker image with the provided examples.

Example api call:

curl --location --request POST 'https://api.banana.dev/start/v4/' \
--header 'Content-Type: application/json' \
--data-raw '{
    "apiKey": "d3a8262d-25e2-4393-ac03-b8e888cf8d8e",
    "modelKey": "5f5051db-1249-4530-be36-a2dc95b0746b",
    "modelInputs" : {
        "image": "UklGRiwAAABXRUJQVlA4ICAAAABQAQCdASoBAAEAAkA4JQBOgCgAAP76eV8CStWlVemeAA=="
    }
}'

Where image is base64-encoded image file.

Example output:

{
    "id": "a46d75e8-e9c9-474b-a939-3fd8e849ceec",
    "message": "",
    "created": 1679143667,
    "apiVersion": "January 11, 2023",
    "callID": "",
    "finished": true,
    "modelOutputs": [
        {
            "class": "111"
        }
    ]
}

The API is fully working, you can check it manually, via Postman.