yoloscr: Live Screen Object Detection

A simple and powerful command-line tool to run any Ultralytics YOLO model for live object detection directly on a region of your screen.

It's perfect for developers, researchers, and hobbyists who want to quickly apply object detection to videos, games, or any application window without complex code.

Features

• Live Detection: Run YOLO models in real-time on your screen.
• Customizable Region: Specify any part of your screen for detection.
• Model Agnostic: Works with any *.pt model file from the Ultralytics ecosystem.
• Highly Configurable: Control confidence, FPS, window visibility, and logging via simple command-line flags.
• Cross-Platform: Works on macOS, Windows, and Linux (with OS-specific setup).

Installation

You can install yoloscr directly from PyPI:

pip install yoloscr

Crucial Operating System Setup

Because this tool needs to see your screen, you may need to grant permissions first.

macOS (Required)

On macOS, you must grant screen recording permissions to your terminal application.

1. Open System Settings.
2. Go to Privacy & Security > Screen Recording.
3. Find your terminal application (e.g., Terminal, iTerm2, Visual Studio Code) in the list.
4. Enable the switch next to it.
5. You will be prompted to "Quit & Reopen" the application. This is necessary for the changes to take effect.

Windows & Linux

On most Windows and Linux systems, no special permissions are required. It should work out of the box.

Usage

The tool is run from your terminal. The basic command requires a YOLO model file.

# Use a pre-trained model downloaded automatically
yoloscr --model yolov8n.pt

# Use your own fine-tuned model
yoloscr --model "runs/detect/train/weights/best.pt"

Command-Line Arguments

You can customize the detection with the following arguments:

Practical Examples

1. Detect on a 400x400 area at the top-left of the screen:

yoloscr --model yolov8n.pt --region 0 0 400 400

2. Use your custom model with high confidence and see logs: Drag the preview window out of the capture area to avoid the "mirror" effect.

yoloscr --model path/to/my/best.pt --conf 0.75 --verbose

3. Run headless (no preview) and just log detections in the terminal: This is useful for running automated scripts.

yoloscr --model yolov8n.pt --hide-window --verbose

License

This project is distributed under the MIT License. See the LICENSE file for more information.