isekA.I — AI vs Human Anime Illustration Detector

A machine learning-based web application designed to classify anime illustrations into two categories: Human-made or AI-generated. The system utilizes a fine-tuned EfficientNetB0 deep learning model for classification.

This project was developed in fulfillment of the Final Project / Final Semester Examination (UAS) requirements for the Perilaku Manajerial course.

• Modern User Interface — Clean and minimal design to facilitate user interaction.
• EfficientNetB0 Model — Fine-tuned on a dataset of 6,485 anime illustrations for accurate classification.
• Drag & Drop File Upload — Asynchronous image upload with instant client-side preview.
• Real-time Inference — Rapid classification with visual probability distribution.
• Responsive Design — Adaptive layout supporting desktop, tablet, and mobile environments.
• Session-based History — Tracks recent classification results within the active browser session.
• Data Export — Functionality to download or copy prediction metrics for external reporting.

isekA.I/
├── app.py # Flask application and routing definitions
├── utils.py # Image preprocessing and model inference utilities
├── requirements.txt # Python environment dependencies
├── README.md # Project documentation
├── best_EfficientNet_finetuned.keras # Pre-trained deep learning model weights
│
├── templates/
│ └── index.html # Main HTML structure
│
├── static/
│ ├── css/
│ │ └── style.css # Cascading Style Sheets
│ ├── js/
│ │ └── script.js # Client-side application logic
│ ├── uploads/ # Auto-generated directory for temporary files
│ └── img/ # Static image assets
│
└── assets/ # Documentation assets

• Python 3.8 or higher
• pip (Python package installer)

1. Navigate to the project directory:

   cd isekA.I

2. Initialize a virtual environment (Recommended):

   • Windows:

     python -m venv .venv
     .venv\Scripts\activate

   • macOS/Linux:

     python3 -m venv .venv
     source .venv/bin/activate

3. Install required dependencies:

   pip install -r requirements.txt

4. Launch the development server:

   python app.py

5. Access the application: Navigate to http://127.0.0.1:5000 in a standard web browser.

1. Image Ingestion: Convert input image to RGB color space.
2. Resizing: Scale image to standard resolution of 224 ×ばつ 224 px.
3. Array Conversion: Transform image data into a float32 NumPy array.
4. Dimension Expansion: Expand tensor dimensions to match batch requirements (1, 224, 224, 3).
5. Normalization: Apply EfficientNet-specific preprocessing to scale pixel values.
6. Inference: Evaluate the tensor through the fine-tuned Keras model utilizing a Sigmoid activation output.
7. Classification Decision: Apply a decision threshold at 0.5 (Score $\ge$ 0.5 indicates AI Generated; Score < 0.5 indicates Human Art).

Renders the primary application interface.

Processes the submitted image file and returns classification statistics.

• Content-Type: multipart/form-data
• Field: file (Image payload)

JSON Response Format:

{
 "error": false,
 "prediction": "AI Generated",
 "emoji": "🤖",
 "confidence": 96.84,
 "ai_probability": 96.84,
 "human_probability": 3.16,
 "processing_time": 0.245,
 "raw_score": 0.9684,
 "image_url": "/static/uploads/filename.png"
}

• Course: Perilaku Manajerial
• Project Status: Tugas Akhir / Ujian Akhir Semester (UAS)