🧠 YOLO_Team_Labeler

A centralized, multi-client image annotation system with server-side YOLO inference and training.

Designed for collaborative dataset creation without data duplication. This platform enables multiple annotators to work in parallel on a shared image pool while a single AI model runs on the server, learns continuously, and serves predictions to all clients.

Core Philosophy: Time is more valuable than compute. By combining human supervision with AI assistance, you can label thousands of images in hours, not days.

---

📌 The Problem vs. The Solution

The Old Way ❌	The YOLO Team Labeler Way ✅
Data Silos: Images copied to every laptop.	Centralized: Images/Labels exist only on the server.
Conflicts: Labels scattered & duplicate work.	Parallelism: Automatic image assignment & locking.
Static: No feedback loop while labeling.	Active Learning: Server trains models on the fly.
Heavy: Users need powerful GPUs.	Lightweight: Clients are thin; Server does the heavy lifting.

---

🔄 The "Active Learning" Workflow

This tool accelerates dataset creation for embedded models through an iterative loop:

1. Phase 1: Cold Start (Manual)
   • Label the first batch of images (approx. 5% of dataset) manually without AI assistance.
2. Phase 2: Initial Training
   • Trigger a training run directly on the server using this small labeled dataset.
3. Phase 3: AI-Assisted Labeling
   • Switch the server to use the newly trained model.
   • The Speed Boost: Clients now receive pre-predicted bounding boxes. Users only need to verify or correct the boxes (Ctrl + Drag) rather than drawing from scratch.
4. Phase 4: Iterative Improvement
   • Retrain the model after every ~200 new images.
   • As model accuracy hits >50%, labeling speed doubles.
   • As accuracy hits >80%, users become purely supervisors, enabling rapid processing of massive datasets.

---

🏗️ System Architecture

🧩 Components

Component	Responsibility	Tech Stack
Server (server.py)	The Brain. Serves images, prevents conflicts, stores labels, runs YOLO inference, and trains models.	FastAPI, Ultralytics, Pillow, Uvicorn
Client (client.py)	The Hands. Connects to server, displays images, captures user input, and requests AI predictions.	Tkinter, Requests, Pillow

---

📸 Application Gallery

1. Raw Input	2. AI Prediction
User receives raw image	Server sends Auto-Labels

3. In-App Training	4. Data Export
Train directly from GUI	Export to YOLO format

---

🚀 Getting Started

1. Installation

Create a virtual environment and install dependencies:

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Start the Server

Run the server on the machine that holds the images:

python3 server.py

• GUI Action: Select your images folder.
• GUI Action: Select a base model (e.g., yolov8n.pt).
• GUI Action: Click "Start Server".

3. Start the Client(s)

Run the client on any machine on the same network:

python3 client.py

• Login: Enter the Server IP (e.g., 192.168.1.X) and your Username.

---

🎮 Controls & Shortcuts

Context	Action	Function
Draw	Left-Click + Drag	Create a new bounding box.
Edit	Ctrl + Drag	Move an existing box.
Resize	Ctrl + Drag Corner	Resize the specific corner.
Delete	Right-Click	Delete the box under the cursor.
Nav	D	Submit labels & load Next image.
Nav	A	Go to Previous image.
Class	W / S	Cycle through object classes Up / Down.

---

⚙️ Advanced Details

Intelligent Label Storage

Labels are stored in standard YOLO .txt format in a labels_collected/ directory on the server.

<class_index> <x_center> <y_center> <width> <height>

Dataset Export

The server includes a one-click export feature that structures your data for external training (e.g., on a cloud GPU cluster):

dataset_export/
├── images/
├── labels/
└── data.yaml  # Ready for Ultralytics/Roboflow

🤝 Contributing

Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/Feature)
3. Commit your Changes (git commit -m 'Add some Feature')
4. Push to the Branch (git push origin feature/Feature)
5. Open a Pull Request

📄 License

Distributed under the MIT License. See LICENSE for more information.

---

Built by Chetan Parihar