Skip to content

setup.md

Latest commit

 

History

History
86 lines (60 loc) · 1.97 KB

setup.md

File metadata and controls

86 lines (60 loc) · 1.97 KB
Raw

Setup Guide

Complete setup instructions for running IRIS locally or on a cluster.

Prerequisites

  • Python 3.11+
  • uv package manager
  • GPU server with CUDA support (for inference)

Installation

git clone https://github.com/EPFL-AI-Team/IRIS
cd IRIS
uv sync

Local Development

The system requires two processes: a server (GPU) and a client (camera/web interface).

Terminal 1: Server

uv run iris-server

Server runs on http://localhost:8005 by default.

Terminal 2: Client

uv run iris-client

Web interface available at http://localhost:8006

Raspberry Pi Setup

The idea of this setup is to ensure the Pi can communicate to the cluster, if the server script is running there.

Generate SSL Certificate (First Time Only, for SCITAS or RCP usage)

For HTTPS streaming from Raspberry Pi:

mkdir -p ~/iris-certs
cd ~/iris-certs
openssl req -x509 -newkey rsa:4096 -nodes \
  -keyout key.pem -out cert.pem -days 365 \
  -subj "/C=US/ST=State/L=City/O=Organization/CN=$(hostname -I | awk '{print $1}')"

Enable HTTPS

In config.yaml, set:

client:
  web:
    use_ssl: true
    cert_dir: "~/iris-certs"

Start Client

uv run iris-client

Connect via https://<raspberry-pi-ip>:8006

Configuration

Two config files are provided in the configs/ directory:

  • configs/config.yaml - Local/Mac development
  • configs/config.rcp.yaml - EPFL cluster deployment

Using arguments with the scripts you're running, you can specify which config you want to use. By default, the usual configs/config.yaml is used.

Key settings:

  • server.model_id - Model checkpoint path
  • server.vlm_hardware - Hardware profile (mac, v100, a100)
  • server.batch_inference.enabled - Enable batch processing

See Configuration Reference for all options.

Cluster Deployment

For running on EPFL clusters (Izar/RCP), see Cluster Setup Guide.