YOLO Streaming Demo (Server + Local Webcam Client + Browser Client)

Overview

• server.py: FastAPI WebSocket server that runs a YOLO model and returns detections for each frame.
• client.py: Python desktop client that streams your laptop webcam to the server and draws detections locally.
• web_client.html: Browser client (works on desktop and iPhone Safari) that streams camera frames to the server and can draw returned boxes.

Quickstart

1. Install requirements (prefer a virtualenv):

   pip install -r requirements.txt

2. Run the server (downloads YOLO weights on first run):

   python -m uvicorn server:app --host 0.0.0.0 --port 8000

3. Run the desktop client (on the same machine or a different one):

   • Edit SERVER_WS_URL in client.py to point to your server (e.g., ws://<server-ip>:8000/ws)

   python client.py

   • Press q to quit.

4. Try the browser client:

   • Open web_client.html in a browser over HTTPS or via python -m http.server and visit http://localhost:8000 (adjust for your static server).
   • Click Connect (allow camera). It will stream frames to ws://127.0.0.1:8000/ws by default—change the field to your server URL if remote.

Notes

• Protocol: client sends TEXT JSON meta then BINARY JPEG; server replies TEXT JSON with detections.
• Tweak JPEG_QUALITY and TARGET_FPS on the clients for bandwidth/latency tradeoffs.
• For GPU inference, install CUDA-compatible PyTorch and Ultralytics as per their docs.
• Security: this is a demo—add authentication, rate limiting, and TLS for production.
• iPhone: use web_client.html in Mobile Safari; frame sizes and FPS should be lowered for stability.

Files

• server.py
• client.py
• web_client.html
• requirements.txt