A focused interface
for traffic detection.
Upload a dashcam frame, run YOLOv8 inference, and review every detected object and traffic violation in one workspace.
Detection classes
23
vehicles, signs, signals, pedestrians
Validation mAP50
92.7%
on the 1,470-image validation split
Detection vocabulary
0 classes · YOLOv8
Real detection workspace
Upload a frame, run YOLOv8 inference, inspect every object, and export the annotated result without leaving the app.
Structured violation engine
Detections are cross-checked through rule logic so results read like an operational review, not a demo overlay.
Persistent review trail
Every run can be saved to local history, revisited in analytics, and exported for handoff or audit.
What it
does today.
A custom-trained YOLOv8s detector feeds a rule engine that flags likely violations, and every run can be saved for later review.
Detection classes
0
vehicles, signs, signals, pedestrians
Model inference
~0ms
per image on an Apple M3 Pro GPU; about 87 ms on CPU (the free live demo runs on CPU)
Dataset coverage
0,254
labeled training images
Workspace pages
0
detection, history, analytics, settings
Upload
Drop a JPG, PNG, or WebP frame into the detection workspace.
Review
Inspect annotated output, confidence values, and rule-level findings.
Track
Saved analyses appear inside history and feed the analytics layer.
Tune
Preferences control autosave behavior and completion notifications.
Three parts: overview, system, workspace.
This page gives the overview, the system page explains the model and rule engine, and the workspace pages run detections and keep a history of past results.
Open the workspace.
Run a detection.
Built to be used, not just viewed. Open the detection workspace and run your first scan in under a minute.