A computer vision web app built with ViT (Vision Transformer) and an ANN classifier to detect oral diseases from mouth images.

Developed for the APS360 course project at the University of Toronto.

This Streamlit-powered app allows users to upload images of mouths or teeth and receive predictions for the following oral diseases:

• Calculus
• Caries
• Gingivitis
• Hypodontia
• Ulcer
• Tooth Discoloration

The pipeline uses:

• 🔎 ViT (pretrained on ImageNet) to extract deep features
• 🧠 ANN classifier trained on extracted ViT embeddings
• 🌐 Deployed using Streamlit Cloud

1. Clone the repo

   git clone https://github.com/ymoola/APS360_Project.git
   cd APS360_Project/oral_disease_app
   

2. Install Dependencies

   pip install -r requirements.txt
   

3. Run App

   streamlit run app.py
   

💡 Make sure the model file model_VIT_bs32_lr0.0001_epoch24 is present in the root of the repo or update the path in predict.py.

App is deployed here: https://oral-disease.streamlit.app

ViT Base (Patch16, 224) from timm is used for feature extraction. A lightweight fully connected layer (ANN) classifies extracted embeddings into 6 oral disease categories.

APS360_Project/
├── oral_disease_app/
│   ├── app.py
│   ├── model/
│   │   ├── vit_feature_extractor.py
│   │   ├── ann_classifier.py
│   │   └── model_VIT_bs32_lr0.0001_epoch24
│   ├── utils/
│   │   └── predict.py
│   ├── requirements.txt
│   └── README.md