The model

The AI behind the scan.

EpivaScan runs on an image-classification model that looks at a photo of skin and scores it against a set of conditions. This page is where its details will live.

Model in development Model card pending

Planned pipeline

What happens to a photo.

  1. 01

    Photo

    An image of the skin area, taken in the app or uploaded.

  2. 02

    Prepare

    The image is resized and normalised into the format the model expects.

  3. 03

    Classify

    The model scores the image against each condition it was trained on.

  4. 04

    Report

    Scores are shown as percentages, highest first.

Model card

The facts, once they exist.

Everything below is filled in when the model is published. Nothing here is a claim yet.

View on Hugging Face
huggingface.co · model
Hosting
Hugging Face · link pending
Task
Image classification
Architecture
To be published
Training data
To be published
Conditions covered
To be published
Validation results
To be published
Results by skin tone
To be published
Version
Unreleased

Conditions covered

What it will recognise.

The list of conditions is being finalised alongside the model. Each one will be listed here with a plain-language description.

Condition 01 Condition 02 Condition 03 Condition 04 Condition 05 Condition 06 …

Transparency

What we plan to publish.

A tool that looks at people’s skin should show its working. These are the things this page is set up to report.

Results across the full range of skin tones.

  • Accuracy for each conditionHow often the top match is right, condition by condition.
    Pending
  • Performance by skin toneWhether results hold up on lighter and darker skin alike.
    Pending
  • Where the data came fromThe image sources used for training and testing.
    Pending
  • Known limitationsConditions, body areas and photo types the model handles poorly.
    Pending

See where the model plugs in.

The web demo already has the capture and results screens. It lights up when the model is connected.

Open the demo