SkinConditionClassifier

Project Overview

SkinConditionClassifier is an Android application that uses Firebase AutoML to classify skin conditions from images. The app allows users to select an image from their device, crop it, and then processes the image using a local TFLite model to predict the skin condition. The model can classify 15 different skin conditions, including Eczema, Acne, Melanoma, and more.

Motivation

The project aims to provide a quick, on-device tool for preliminary skin condition assessment. By leveraging machine learning, it helps users identify potential skin issues without requiring immediate professional consultation, thus raising awareness and encouraging timely medical advice.

Features

• Image Selection & Cropping: Users can select an image from their device and crop it for better analysis.
• On-Device Classification: Utilizes a local TFLite model for fast, offline skin condition classification.
• Multiple Condition Detection: Classifies 15 different skin conditions with confidence scores.
• User-Friendly Interface: Simple UI with an image preview and clear classification results.

Architecture & Diagrams

System Architecture

graph TB
    subgraph Android App
        UI[User Interface]
        ImagePicker[Image Picker]
        Cropper[Image Cropper]
        MLKit[Firebase ML Kit]
        TFLite[TFLite Model]
    end
    
    subgraph Firebase Services
        AutoML[Firebase AutoML]
        ModelManager[Model Manager]
    end
    
    UI --> ImagePicker
    ImagePicker --> Cropper
    Cropper --> MLKit
    MLKit --> TFLite
    TFLite --> Results[Classification Results]
    
    AutoML -.-> ModelManager
    ModelManager -.-> TFLite

Loading

Data Flow

sequenceDiagram
    participant User
    participant App
    participant MLKit
    participant Model
    
    User->>App: Select Image
    App->>App: Crop Image
    App->>MLKit: Process Image
    MLKit->>Model: Run Inference
    Model->>MLKit: Return Predictions
    MLKit->>App: Format Results
    App->>User: Display Classification

Loading

Component Interaction

classDiagram
    class MainActivity {
        +onCreate()
        +onActivityResult()
        +setLabelerFromLocalModel()
        +processImageLabeler()
    }
    
    class FirebaseML {
        +getInstance()
        +getOnDeviceAutoMLImageLabeler()
    }
    
    class ImageProcessor {
        +cropImage()
        +prepareImage()
    }
    
    class ModelManager {
        +downloadModel()
        +updateModel()
    }
    
    MainActivity --> FirebaseML
    MainActivity --> ImageProcessor
    FirebaseML --> ModelManager

Loading

Components:

• User Interface: Built using Android XML layouts (activity_main.xml).
• Image Selection & Cropping: Uses the android-image-cropper library.
• Firebase AutoML: Integrates Firebase ML Kit for model management and inference.
• TFLite Model: A local model (model.tflite) bundled in the assets folder, with labels defined in dict.txt.

How It Works

1. Image Selection: The user taps the "Select Image" button, which opens the device's image picker.
2. Image Cropping: The selected image is cropped using the CropImage library.
3. Model Inference: The cropped image is processed by the local TFLite model using Firebase AutoML.
4. Classification: The model predicts the skin condition and displays the result with a confidence score.

Setup & Installation

1. Clone the repository.
2. Open the project in Android Studio.
3. Update google-services.json with your Firebase API key.
4. Build and run the application on an Android device or emulator.

Usage

1. Launch the app.
2. Tap "Select Image" to choose an image from your device.
3. Crop the image as needed.
4. View the classification result displayed below the image.

Learnings

• Integration of Firebase ML Kit with Android applications.
• Handling image selection and cropping in Android.
• On-device machine learning using TFLite models.

Possible Improvements

• Remote Model Updates: Implement dynamic model updates from Firebase.
• Enhanced UI: Improve the user interface with better styling and feedback.
• Additional Conditions: Expand the model to classify more skin conditions.
• Offline Support: Ensure robust offline functionality.

License & Credits

This project is licensed under the MIT License. Special thanks to Firebase and the Android community for their tools and libraries.