LiveDepth

LiveDepth is a real-time monocular depth streaming app for macOS.
It uses PyTorch MiDaS models to infer per-pixel depth from a webcam feed (or a synthetic test pattern), and streams results over WebSockets to a browser-based 3D viewer.

It runs locally, requires no cloud connection, and packages into a standalone .app with HTTPS UI and WSS data channel.

---

Features

• Monocular depth estimation in real-time using MiDaS (MiDaS_small, DPT_Hybrid, or DPT_Large).
• Camera feed (FaceTime HD, external USB, or virtual webcams) or synthetic test pattern.
• Streams depth + RGB frames via WebSocket.
• Local HTTPS server hosts the UI viewer.
• Status-bar app bundle for macOS with:
  • Open Viewer
  • Restart Server
  • About panel (with version/build metadata)
  • Quit

---

Requirements

• macOS 11.0+ (Apple Silicon recommended).
• Python 3.10+ (tested with 3.13).
• Homebrew (to install ffmpeg and mkcert).
• Dependencies (install inside a venv):

  pip install opencv-python torch torchvision torchaudio pillow websockets numpy timm pyobjc

---

Running Locally (Source Tree)

1. Clone and set up environment

git clone https://github.com/sylwesterdigital/webcam-monocular-depth
cd webcam-monocular-depth
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

2. Generate HTTPS certificates (first run only)

cd client
mkdir -p certs && cd certs
brew install mkcert nss
mkcert -install
mkcert localhost 127.0.0.1 ::1

3. Test run (without build)

Use test pattern (no camera):

./test.sh

Or use webcam:

./test.sh --camera

This launches:

• HTTPS UI at https://127.0.0.1:8443
• WebSocket server at ws(s)://127.0.0.1:8765/

---

Building macOS .app

We use PyInstaller with a custom LiveDepth.spec.

1. Ensure you have pyinstaller

pip install pyinstaller

2. Create app icon

mkdir -p build/Icon.iconset assets
sips -z 16 16     icon.png --out build/Icon.iconset/icon_16x16.png
sips -z 32 32     icon.png --out build/Icon.iconset/[email protected]
sips -z 32 32     icon.png --out build/Icon.iconset/icon_32x32.png
sips -z 64 64     icon.png --out build/Icon.iconset/[email protected]
sips -z 128 128   icon.png --out build/Icon.iconset/icon_128x128.png
sips -z 256 256   icon.png --out build/Icon.iconset/[email protected]
sips -z 256 256   icon.png --out build/Icon.iconset/icon_256x256.png
sips -z 512 512   icon.png --out build/Icon.iconset/[email protected]
sips -z 512 512   icon.png --out build/Icon.iconset/icon_512x512.png
sips -z 1024 1024 icon.png --out build/Icon.iconset/[email protected]
iconutil -c icns build/Icon.iconset -o assets/LiveDepth.icns

3. Build

pyinstaller LiveDepth.spec --noconfirm

The result is in:

dist/LiveDepth.app

Move it to /Applications if desired.

---

Usage

• Launch LiveDepth.app from Finder.

• A status-bar icon appears (LiveDepth).

• Menu options:

  • About LiveDepth — shows version/build info
  • Open Viewer — opens the browser client
  • Restart Server — restarts depth inference
  • Quit LiveDepth — exits

---

Configuration

Set via environment variables:

• WEBCAM_INDEX=0 — select camera by index
• WEBCAM_NAME="FaceTime HD Camera" — select camera by name (macOS/AVFoundation)
• MODEL_TYPE=MiDaS_small|DPT_Hybrid|DPT_Large
• TEST_PATTERN=1 — use synthetic scene (for debugging)
• TARGET_WIDTH=640 — resize input width
• STRIDE=2 — subsample for performance/bandwidth

---

License

© 2025 Sylwester Mielniczuk — WORKWORK.FUN LTD (UK). All rights reserved.