RobfarmerModel Docker Setup

This guide explains how to build and run the RobfarmerModel using Docker.

Prerequisites

• Docker installed on your system
• NVIDIA Docker runtime (for GPU support)
• Copy the Robofarmer-II directory from elvex.sintef.no:/data/robofarmer to your home directory

Building the Docker Image

Build the Docker image with the following command:

podman build -t robofarmer:latest .

Note: The build process will:

• Use NVIDIA PyTorch base image with CUDA support
• Install system dependencies and build tools
• Set up Anaconda environment
• Clone the repository and install dependencies
• Build OpenCV with CUDA support
• Extract the dataset from Robofarmer-II.tar.gz

Running the Container

Basic Run Command

podman run -it --rm robofarmer:latest

With GPU Support

podman run -it --rm --gpus all robofarmer:latest

With Volume Mounting (for data persistence)

podman run -it --rm --device nvidia.com/gpu=0 \
  -v $(pwd)/Robofarmer-II:/app/data/datasets/Robofarmer-II  \
  robofarmer:latest

Interactive Shell Access (Recommended for running all scripts)

podman run -dit --shm-size=24g --device nvidia.com/gpu=0 --name robofarmer -v $HOME/Robofarmer-II:/app/data/datasets/Robofarmer-II bash

Enter the container

podman attach robofarmer

Working Directory

The container's working directory is /app, which contains:

• /app/src - The cloned RobfarmerModel repository
• /app/data - Dataset directory
• /app/opencv - OpenCV source code

Conda Environment

The container uses a conda environment called samclip. To activate it within the container:

conda activate robrobofarmer

Common Commands

Once inside the container, you can run the model scripts:

# Activate the environment
conda activate robofarmer

# Navigate to source directory
cd src/interaction_hotspots

How to train and evaluate models

In the interaction_hotspots directory, there are README files explaining how to train and evaluate models and to visualize heatmaps using trained models.