Recently, I built a simple browser-based Face Detection app powered by an embedded AI model using Ultraface ONNX.

The workflow is pretty straightforward:

1. Drag and drop an image
2. Preprocess the image (Image → Canvas → Pixels → Tensor)
3. Run inference using ONNX Runtime Web
4. Execute model to generate detection results (confidence filtering and IoU for NMS)
5. Extract final face bounding boxes
6. Render results with a beautiful UI

In the UI, you can also see:

1. Total number of detected faces in an image
2. File size of the uploaded image (max with 5 MB)
3. Average confidence score, which reflects the overall reliability of detections

Because it runs on ONNX Runtime Web, everything works fully offline — no backend or external API calls needed.

Really excited to explore more around lightweight AI, embedded models, and client-side inference stuff 👀, so fun, really.

Live Demo:

facetecta.alimnfl.com

Result:

#ComputerVision #ONNX #MachineLearning #ArtificialIntelligence