Object detection
Run a bundled SSD MobileNet detector or bring your own Core ML model for labeled detections.
Local Core ML vision for macOS
Mac Vision Tools
A menu bar workspace for real-time object detection, emotion monitoring, privacy guardrails, and focus sessions from camera or screen capture.
Four workflows
Built for live visual awareness
Face emotion cues
Detect faces first, classify the visible emotion, and keep a short recent history.
Presence threshold
Count visible people and start the screen saver when your configured threshold is reached.
Attention timer
Use native Apple Vision head-pose tracking to count focused time toward a session goal.
Camera or screen
Choose the capture source and display style
Model contracts
Know what to train or drop in
Mac Vision Tools accepts local Core ML models. Use these contracts when replacing a bundled model, training a new one, or choosing whether a file belongs in detection or emotion mode.
SSD MobileNet V2 COCO
Used by Standard Detection and Privacy Guard for live object labels and person counting.
Task
Object detection
Classes
80 COCO object classes. Privacy Guard depends on the
person class being present and named clearly.
View the bundled COCO labels
person, bicycle, car, motorcycle, airplane, bus, train, truck, boat, traffic light, fire hydrant, stop sign, parking meter, bench, bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair drier, toothbrush.
Input
Bundled export uses 320 x 320 image input. Custom detectors can use another size if the Core ML model is accepted by Vision.
Output
Vision object detections, or SSD-style multi-array boxes, scores, class ids, and count outputs compatible with the app parser.
Train/drop in
Train a Core ML detector with stable labels, select it in Standard Detection first, then use it for Privacy Guard only if it reliably emits
person.
EmotiEff emotion model
Used by Emotion mode after Apple Vision finds and crops a visible face.
Task
Face emotion classification
Classes
7 bundled labels:
Angry, Disgust, Fear, Happy, Neutral, Sad, and Surprise.
Input
Face crop supplied by Apple Vision and scaled to the bundled model's 224 x 224 image input. Train replacement classifiers on face crops, not full camera frames.
Output
Predicted class label and confidences compatible with Vision classification results.
Train/drop in
Export an image classifier as Core ML. Keep labels equivalent to the bundled emotions so colors, history, and summaries stay meaningful.
Apple Vision
Used for face detection, face crops, landmarks, and native focus tracking.
Classes
No trainable app classes. Apple Vision provides system observations instead of a bundled model file.
Drop in
Nothing to replace for Focus mode. The app uses the macOS framework automatically.
Custom Core ML models
Select local model files when experimenting with your own detector or classifier.
Formats
.mlmodel, .mlpackage, or compiled .mlmodelc files and folders selected from your Mac.
Detection
Use Standard Detection to validate labels and boxes. Privacy Guard needs a detector that emits people as
person.
Emotion
Use a classifier with labels that match or clearly map to the 7 bundled emotion labels above.
Licensing
Custom models stay local, and you are responsible for permission to use or redistribute any model you select.
For export details, see the model replacement notes. For bundled notices, see the app's Notices view and repository credits.
On-device by design
Inference stays local
Mac Vision Tools uses Core ML, Vision, AVFoundation, and ScreenCaptureKit on your Mac. Camera and screen frames are processed for live detections and are not saved by the app.
Permissions are explicit: camera capture needs Camera access, and screen capture needs Screen Recording access in macOS System Settings.
The app does not create accounts, run analytics, track users, or send camera or screen content to a server. Custom Core ML models selected by the user stay on the Mac.
Privacy Guard starts the macOS screen saver when the configured person threshold is reached. macOS controls whether the screen saver requires a password.
Open source macOS utility
Build it, run it, adapt it
The app ships with bundled models and also accepts custom Core ML model files for mode-specific experiments.