Cross-platform desktop YOLO inference
Three input sources — images, video files, and RTSP streams — with automatic CUDA acceleration. ROI alarms, fall detection, ByteTrack, and support for any Ultralytics .pt model, ready out of the box.
What it does
Each capability maps to a concrete everyday problem.
Three input sources
Single images, local video files, and live RTSP streams all work directly — no transcoding or extra setup.
Detection and pose
YOLOv8-based detection and pose estimation, outputting class, confidence, and keypoints together.
ROI zone alarms
Draw a zone of interest and trigger an alarm when a target enters. Inverse mode detects absence instead.
Fall detection
Uses pose keypoints to detect falls, useful for elder care and site safety.
ByteTrack tracking
Assigns persistent IDs to targets, with label filtering so you only watch the classes you care about.
Automatic CUDA
Automatically uses GPU inference when an NVIDIA card is present, and falls back to CPU otherwise.
How it works in practice
Not a feature list — a walk through getting something done.
Point it at a camera and see results live
Enter an RTSP URL and inference starts on the live feed, showing boxes, confidence, and frame rate in real time.
Switch between multiple streams to keep an eye on several key positions at once.
When only one zone matters, draw it
ROI supports polygon selection so alarms only fire when a target enters the zone, cutting noise from irrelevant objects.
Inverse mode suits "someone must be present" cases such as monitoring an attended post.
Any Ultralytics .pt works
No code changes — load a trained .pt file and run inference, with class names read from the model.
Together with Easy Dataset and Easy Trainer Station it forms a full chain: prepare data, train, then infer.
Design trade-offs
Deterministic work stays in code; judgement stays with you.
GPU first
Detects CUDA automatically and uses the GPU when available, CPU otherwise.
Install and run
Installers for Windows, macOS, and Linux — no Python environment to set up.
Visual parameters
Adjust confidence thresholds, IOU, and label filters in the UI and see the effect immediately.
Toolchain integration
Shares data and model conventions with Easy Dataset and Easy Trainer Station.
What you might want to know first
Do I need a dedicated GPU?
No. With an NVIDIA card CUDA is enabled automatically; without one it runs on CPU, just slower.
Can it use a camera?
Yes — enter the camera RTSP URL and it pulls the stream directly.
Can I use my own model?
Any .pt model exported from Ultralytics works, and class names are read from the model.
Start using it now
Free to download, runs locally. Your data stays on your machine.
This is a desktop app — open this page on your computer to download.
Other formats
SHA256 checksum
Provided as-is. The app does not collect usage data.