Running Roboflow Inference

Overview

Roboflow is a computer vision platform for building, training, and deploying image models and workflows. This guide sends images from a container on an ALPON X5 AI or ALPON X4 to the Roboflow hosted API and returns inference results. Processing is remote rather than on-device, so latency, data handling, usage limits, and availability depend on the network and your Roboflow plan.

The application workflow is simple: the app is dockerized, the container is deployed to the ALPON via ALPON Cloud, and the deployed application runs continuously on the device. The steps below are identical on ALPON X4 and ALPON X5 AI.

1. 1

   Create a workflow in Roboflow

   Go to app.roboflow.com and sign up or log in. In the left menu, select Workflows → Create a Workflow.

   

   Add a model by clicking Add a Model, then select Object Detection model.

   

   

   Add a model by clicking Browse Pre-trained Models, then select YOLOv8.

   

   

   Save the workflow.

   

   Go to Settings → API Keys and copy your API key.

   

   

   In the workflow view, click Deploy to see the required environment variables — the workspace name and workflow ID. You’ll need them in step 5.

   

2. 2

   Write the Roboflow inference app

   Package your Roboflow client as a Docker container. Create a directory on your build machine with the following Dockerfile and main.py:

   Dockerfile

   FROM python:3.11-slim
   
   WORKDIR /app
   
   RUN pip install --no-cache-dir inference-sdk
   
   COPY main.py .
   
   CMD ["python", "main.py"]

   python · main.py

   # main.py
   import os
   import time
   from inference_sdk import InferenceHTTPClient
   
   API_URL = os.getenv("ROBOFLOW_API_URL", "https://serverless.roboflow.com")
   API_KEY = os.getenv("ROBOFLOW_API_KEY")
   WORKSPACE_NAME = os.getenv("ROBOFLOW_WORKSPACE")
   WORKFLOW_ID = os.getenv("ROBOFLOW_WORKFLOW_ID")
   IMAGE_PATH = os.getenv("IMAGE_PATH", "/data/image.jpg")
   
   if not all([API_KEY, WORKSPACE_NAME, WORKFLOW_ID]):
       raise EnvironmentError("Missing required environment variables.")
   
   client = InferenceHTTPClient(api_url=API_URL, api_key=API_KEY)
   
   while True:
       try:
           result = client.run_workflow(
               workspace_name=WORKSPACE_NAME,
               workflow_id=WORKFLOW_ID,
               images={"image": IMAGE_PATH},
               use_cache=True
           )
           print("Inference result:", result)
       except Exception as e:
           print("Error during inference:", e)
       time.sleep(60)

   Image inputs only

   The Roboflow API supports only image inputs. If you want to use a video or live feed, you must extract frames and send them one by one (edge inference is not available on ALPON X4).

   You may also modify the code to use a USB camera via cv2.VideoCapture(), though this is not officially supported in the cloud-based flow.

3. 3

   Build the container image

   In the folder containing your Dockerfile and main.py, build the image for the device’s arm64 architecture:

   bash · build the image

   docker buildx build --platform linux/arm64 --load -t roboflow-inference .

4. 4

   Add the Roboflow inference container

   Make the roboflow-inference image available to your device by pushing it to your Sixfab Container Registry, so deployments are fast and repeatable. Log in to Sixfab Registry, click + Add Container, and follow the prompts to push the roboflow-inference image.

   Pushing images to the Sixfab Container Registry

   For the full walkthrough of tagging and pushing an image, see Update Containers from the Sixfab Container Registry.

5. 5

   Deploy the container on ALPON

   Open your device in ALPON Cloud, go to the Applications section, and click + Deploy. In the Deploy Container window, use these settings:

   Container Name roboflow-inference (or your preferred name)

   Image The roboflow-inference image and tag you pushed to the Sixfab Container Registry.

   Environment Click + Add More in the Environment section and add the variables in the table below, with the values from your Roboflow workspace and workflow (step 1).

   Key	Value	Description
   ROBOFLOW_API_KEY	Your Roboflow API key	API key from Roboflow Settings → API Keys
   ROBOFLOW_WORKSPACE	Your Roboflow workspace name	From your Roboflow account
   ROBOFLOW_WORKFLOW_ID	Your Roboflow workflow ID	From your Roboflow Workflow configuration
   ROBOFLOW_API_URL	Optional, default https://serverless.roboflow.com	Roboflow API base URL (optional)
   IMAGE_PATH	Path to input image on device	E.g., /data/image.jpg (optional)

   Click + Deploy to launch the Roboflow inference container on the device.

6. 6

   Check the deployment and inference results

   Once the deployment completes, the roboflow-inference container is running on the target ALPON device and sends the image to the Roboflow hosted API every 60 seconds. Check the Applications tab in ALPON Cloud for real-time status, and open the container logs to see each result:

   ALPON Cloud · container logs

   Inference result: [{ ... }]

   If the log shows Missing required environment variables., check the three required variables from step 5; if it shows Error during inference:, confirm the device has internet access and that the image exists at IMAGE_PATH.

Workflows → Create a Workflow in the Roboflow left menu.

Click Add a Model in the workflow editor.

Select Object Detection model.

Click Browse Pre-trained Models.

Select YOLOv8 from the pre-trained models.

Save the workflow.

Open Settings → API Keys.

Copy your Roboflow API key.

The Deploy panel lists the workspace name and workflow ID.