This folder contains the source code for the image-classification function in OpenFaaS. It implements a basic Python function for image classification using Torchvision.

The function file structure is primarily derived from the python3-http-debian template in the OpenFaaS Python Flask templates. Some files have been removed, and the Dockerfile has been simplified. The main function logic can be found in function/handler.py.

Use the help script to build the image from the project's root directory:

$ ./k8s/scripts/build-image.sh openfaas-image-classification none --dockerfile functions/image-classification/Dockerfile

This will generate the image with localhost/openfaas-image-classification:dev tag. To push the image to GitHub Container Registry (after login) with tag ghcr.io/unimib-datai/dfaas-openfaas-image-classification:

$ ./k8s/scripts/build-image.sh openfaas-image-classification push --dockerfile functions/image-classification/Dockerfile

Use the public image to deploy the function to OpenFaaS:

$ faas-cli deploy --image=ghcr.io/unimib-datai/dfaas-openfaas-image-classification:dev --name=mlimage --label dfaas.maxrate=100

This will deploy the function to the local OpenFaaS instance as mlimage. The dfaas.maxrate label is required by the Recalc strategy. You can omit the label if the DFaaS Agent does not use this strategy.

You can also pass the options --env exec_timeout=4s and --env max_inflight=400 to configure the per-request execution timeout and the maximum number of concurrent requests each replica can handle. See docs/timeout.md for more details.

Download example images from the ImageNet dataset. In this case, we use a few samples from github.com/EliSchwartz/imagenet-sample-images:

$ wget https://github.com/EliSchwartz/imagenet-sample-images/blob/master/n01770393_scorpion.JPEG?raw=true -O scorpion.jpeg
$ wget https://github.com/EliSchwartz/imagenet-sample-images/blob/master/n01616318_vulture.JPEG?raw=true -O vulture.jpeg
$ wget https://github.com/EliSchwartz/imagenet-sample-images/blob/master/n01770393_scorpion.JPEG?raw=true -O scorpion.jpeg

Then, use curl to send the requests:

$ curl -X POST --data-binary "@path/file.jpg" -H "Content-Type: image/jpeg" http://127.0.0.1:30080/function/mlimage
# Example output:
content-length: 58
content-type: application/json
date: Wed, 01 Oct 2025 14:15:56 GMT
server: waitress
x-call-id: 16eef834-f5d2-4b73-ac62-506fcc3e598a
x-duration-seconds: 0.286761
x-served-by: openfaas-ce/0.27.12
x-start-time: 1759328156681799394
x-server: IP

[{"class": "scorpion", "probability": 0.9953627586364746}]

Note that the endpoint is http://127.0.0.1:30080, the HAProxy public port, and not directly the OpenFaaS Gateway. The IP address should be adjusted to match your local OpenFaaS instance.