Precise Live Segmentation

Real-time object detection from your camera where every object gets a pixel-accurate mask tracing its outline — not just a box. Runs fully on your own machine (no cloud), with a clean web control panel.

• YOLO Seg mode — recognizes 80 everyday objects (people, pets, phones, cups, cars…) with name + confidence, tracked across frames for smooth, stable masks
• FastSAM mode — masks everything in view, even objects it can't name
• Works on macOS (Apple GPU), Windows and Linux (NVIDIA GPU if available, otherwise CPU)

Install (one time, ~5 minutes)

You need Python 3.10+ installed. Then open a terminal in this folder and run:

macOS / Linux

python3 -m venv .venv
.venv/bin/pip install -r requirements.txt

Windows (PowerShell)

py -m venv .venv
.venv\Scripts\pip install -r requirements.txt

That's it. Model weights (~20–45 MB each) download automatically the first time each model is used.

Run

macOS / Linux

.venv/bin/python server.py

Windows

.venv\Scripts\python server.py

Your browser opens http://localhost:8000 with the live feed. First launch may ask for camera permission — allow it (on macOS the permission goes to your terminal app) and the feed appears.

Using the control panel

Control	What it does	Shortcut
Segmentation mode	YOLO Seg (named objects) ↔ FastSAM (mask everything)	M
Model size	Nano / Small / Medium — speed vs precision	1 2 3
Confidence	How sure the model must be before masking something	[ ]
Bounding Boxes	Draw classic boxes in addition to masks	B
Class filter	Tap chips (person, car, dog…) to only detect those; chip colour = mask colour
Snapshot	Saves the current annotated frame as a PNG

The LIVE badge shows the active model and real FPS; "Objects detected" updates every second.

Tips

• Small is the best balance (40+ FPS on an Apple M-series GPU). Medium gives the finest masks, Nano is for slower machines — on a CPU-only machine start with Nano.
• On Windows/Linux with an NVIDIA card, install the CUDA build of PyTorch first for GPU speed: pip install torch --index-url https://download.pytorch.org/whl/cu124
• No camera, or want to test with a file? python server.py --source video.mp4 (an image works too).
• Port already in use? python server.py --port 8080

Troubleshooting

• "Camera unavailable" in the feed — close other apps using the camera (Zoom, FaceTime…). macOS: System Settings → Privacy & Security → Camera → allow your terminal. The server keeps retrying every few seconds.
• Slow / laggy — press 1 for the Nano model, or raise confidence with ].
• Browser didn't open — go to http://localhost:8000 manually.

Desktop window (alternative, no browser)

.venv/bin/python app.py        # Windows: .venv\Scripts\python app.py

Same models in a plain OpenCV window. Keys: q quit, m mode, 1/2/3 size, [ ] confidence, b boxes.