Project Info
- Role: Full Stack AI Developer
- Date: June 2025
-
Tech Stack:
Python 3.8+ YOLOv8 FastAPI Tkinter (GUI) Pillow
YOLO Team Labeler
Collaborative, Multi-Client Image Annotation with Active Learning
The Problem
Traditional labeling suffers from data silos (images copied to USBS), conflicting file versions, and requires heavy hardware for every annotator.
The Solution
A centralized Server (The Brain) handles storage and training, while lightweight Clients (The Hands) simply connect to label.
Core Features
- Active Learning Loop: The system trains itself as you label. After labeling ~5% of data, the server trains a model to auto-annotate the rest.
- Smart Workflow: Users switch from "drawing" boxes to "verifying" AI predictions (Ctrl+Drag), doubling speed.
- Conflict Resolution: Automatic image locking ensures no two users label the same image simultaneously.
System Walkthrough
1 Server Setup & Connection
Setting up the central brain and connecting clients via local IP.
Server Dashboard
Client Connection
2 The Labeling Interface
The lightweight GUI for annotators. Images are streamed from the server.
3 Active Learning in Action
How the system uses AI to reduce human workload.
The AI model suggests bounding boxes (Pink). Early in training, these may overlap or be slightly off.
The user quickly adjusts the boxes. In this frame, 4 boxes were auto-correct, saving 50% effort.
4 Training & Export
Triggering training runs from the GUI and exporting the final YOLO-format dataset.
