46 lines (34 loc) · 1.69 KB

Vision Hub – Research Demo Platform

This project lets you deploy and visualize research demos in Docker containers, with a modern web interface.

Prerequisites

Docker & Docker Compose
webcam (V4L2-compatible is a bonus)
NVIDIA GPU + drivers + nvidia-docker for CUDA acceleration (optional, but recommended for best performance)

Quick Start

Clone the repo and go to the vision-hub directory.
Copy .env.example to .env and adjust variables as needed.
Launch the stack:
```
docker compose up --build
```
Access the interfaces:
- Astro frontend: http://localhost:4321
- Web API: http://localhost:8080
- RTSP: rtsp://localhost:8554/cam

Main Environment Variables

See .env.example for the full list. Key variables:

CAMERA_DEVICE: Path to the local camera (e.g. /dev/video0 or /dev/v4l/by-id/...)
RTSP_URL: RTSP stream URL to use (default: rtsp://mediamtx:8554/cam)
MODEL: YOLO model to use (e.g. yolov8n.pt)
IMG_SIZE: Input resolution for inference (e.g. 640)
JPEG_QUALITY: JPEG quality for MJPEG stream (e.g. 80)
FORCE_CPU: Set to 1 to force CPU usage even if a GPU is available
CONF, IOU, MAX_DET: Detection thresholds

Usage Notes

Besides global variables, there's always default ones, these are they I use on my own computer
For GPU acceleration, ensure Docker is configured with the correct runtime (--gpus all or devices: - nvidia.com/gpu=all).
Startup logs will indicate whether GPU or CPU is used.

Custom Demos

Add your own models or demos by creating a dedicated folder and adapting the Dockerfile + app.py.

For questions or contributions, ask me directly at tom.burellier@associated.ltu.se