ProLabeler

ProLabeler v1.1

Smart, automatic video-to-dataset tool for YOLO models (ONNX format)

ProLabeler turns any video (local or YouTube) into a clean, ready-to-train YOLO dataset.
Supports multi-model ensemble detection, GPU acceleration, and automatic label formatting.

To Do

• Add multi-class support
• Video Preview with OpenCV VideoWriter
• Drag and Drop support

---

Features

• Works with YOLOv8 ONNX models (shape 1×5×8400)
• Input from MP4 or YouTube links
• GPU acceleration (CUDA) with automatic CPU fallback
• Ensemble detection with box merging
• Automatic label scaling and aspect-safe resizing
• Clean GUI built with Tkinter + Sun Valley ttk
• Save & load settings (config.json)
• Auto Skips simmilar frames

---

Quick Start

Option 1: Use the compiled .exe

Just download the latest release — no installation needed.

Option 2: Run from source

Prerequisites

• Python 3.10+
• pip and venv installed
• (Optional) NVIDIA GPU + CUDA 12 Toolkit for GPU acceleration
• FFMPEG is required, but it is bundled with this program

Create and activate a virtual environment

python -m venv pro

Windows

pro\Scripts\activate

macOS / Linux

source pro/bin/activate

CPU Version

   pip install .[cpu]

GPU Version (requires CUDA)

   pip install .[gpu]

GPU mode uses onnxruntime-gpu for better performance. But you need CUDA 12.x installed

Launch the GUI

python App.py

Or use the CLI version

Run detections directly via terminal:

python Main.py --video path/to/video.mp4 \
               --models path/to/model1.onnx path/to/model2.onnx \
               --out ./output \
               --conf 0.5 \
               --iou 0.45 \
               --frame-step 1 \
               --merge-iou 0.6

Example (YouTube video):

python Main.py --youtube "https://www.youtube.com/watch?v=dQw4w9WgXcQ" \
               --yt-res 1080p \
               --models ./models/my_model.onnx \
               --out ./output \
               --conf 0.5

Add --cpu to force CPU mode.