README.md

---
language: en
license: mit
library_name: keras
tags:
  - image-classification
  - vision-transformer
  - tensorflow
  - keras
  - medical-imaging
pipeline_tag: image-classification
---

# SquintVue ViT Inference Endpoint

This repository contains the deployed Vision Transformer inference model for SquintVue.

## Model
- File: `strabismus_vit.keras`
- Framework: TensorFlow / Keras
- Backbone: ViT Base Patch16 224
- Classes:
  - ESOTROPIA
  - EXOTROPIA
  - HYPERTROPIA
  - HYPOTROPIA
  - UBIPR_Healthy_Eyes

## Input
JSON request body:

{
  "inputs": {
    "image_base64": "BASE64_ENCODED_IMAGE"
  }
}
## Output
{
  "predicted_class": "UBIPR_Healthy_Eyes",
  "confidence": 0.9631,
  "confidence_percentage": 96.31,
  "class_probabilities": [
    {
      "label": "ESOTROPIA",
      "probability": 0.0086,
      "percentage": 0.86
    }
  ],
  "model_input_size": [224, 224],
  "xai_ready": false
}