PrefNet Studio

A hardware-accelerated, lightweight web application for training binary classification neural networks on CLIP image feature embeddings.

---

Quick Start

Windows (Recommended)

Download the repository by running the following command in the terminal:

git clone https://github.com/Dominik7272/PrefNet-Studio.git

Run the Start.bat launcher in order to start PrefNet Studio. On its first run, it will automatically:

1. Create a local Python virtual environment (.venv)
2. Install all required dependencies (including PyTorch, Flask, and Transformers)
3. Launch the local web server

Manual Launch

If you prefer not to use the batch file, you can set it up manually:

1. Install Dependencies:

   pip install -r requirements.txt

2. Run the App:

   python server.py

3. Open the Interface

Navigate your browser to: http://localhost:8100

---

🛠️ Features Overview

1. Dashboard & Workspaces: Create or load modular workspace configurations which store setting parameters, classification profiles, model paths, batch automated scripts, and recents histories.
2. Dataset Management: Setup distinct classification profiles with customized labels (e.g. Liked/Disliked vs A/B classes).
3. Interactive Keyboard Labeling: Stream local unlabelled image queues with responsive fit-to-screen controls. Classify items instantly using keyboard keys:
   • → (Right Arrow) to label as Class A
   • ← (Left Arrow) to label as Class B
   • ↑ (Up Arrow) to Skip
   • ↓ (Down Arrow) to Undo last labeling
4. Neural Net Training: Tune linear parameters layer weights. View training logs output streams and live-drawn custom SVG coordinate line charts plotting Loss and Accuracy curves in real-time.
5. Bulk Predictions / Inference: Select any trained model weights to categorize folders of images. Filter or sort prediction grids by confidence metrics, and inspect detail cards with manual override capability. Export predictions lists directly as CSV files.
6. Automated Batch Jobs: Script backgrounds jobs to watch target inputs folders and sort new files into class subdirectories based on classification confidence thresholds.

---

Project Structure

PrefNet-Studio/
├── backend/
│   └── app.py          # Flask REST API + Background threads runners
├── frontend/
│   ├── dist/           # Compiled React static resources (served by Flask)
│   ├── src/
│   │   ├── App.jsx     # Main React SPA component
│   │   ├── index.css   # Custom Slate & Indigo Design system styles
│   │   └── main.jsx    # React mounting point
│   ├── package.json
│   └── vite.config.js  # Vite dev server proxies
├── README.md           # Instructions documentation
├── requirements.txt    # Application dependency list
├── server.py           # Unified application boot launcher
└── Start.bat           # Quiet Windows launcher script (creates venv)

---

Development

If you want to make changes to the frontend React source code:

1. Navigate to the frontend/ folder:

   cd frontend

2. Run Vite hot-reload dev server:

   npm run dev

3. Open http://localhost:5173. Any API requests to /api/* will automatically be proxied to the Flask server running on port 8100.
4. Rebuild static assets for production deployment:

   npm run build