Quick Start

This page walks you through installing ATLAS and running your first annotation session. For a deeper look at how to configure ATLAS for your own data, see Configuration Guide.

Prerequisites

  • Python ≥ 3.10

  • conda (recommended for environment isolation)

Installation

Clone the repository and install ATLAS in editable mode inside a fresh conda environment:

# Clone the repository
git clone git@github.com:TUWIEN-ASL/ATLAS-tuwienasl.git
cd ATLAS-tuwienasl

# Create and activate a conda environment
conda create -n gui python=3.10
conda activate gui

# Install ATLAS
pip install -e .

To additionally install the dependencies needed to build this documentation locally, run:

pip install -e ".[docs]"

Running the GUI

Launch the annotation tool by passing one of the example configuration files to the GUI module:

python -m atlas_gui.gui --config config/rlds.yaml

The config/ folder contains ready-to-use configurations for each supported dataset format (REASSEMBLE, RLDS, ROS bags, video, frames). Pick the one that matches your data, copy it, and edit it to point at your own files.

Important

Annotations made on a REASSEMBLE dataset with annotation_storage: h5 are written in place, overwriting the original HDF5 file. Use annotation_storage: json if you would prefer ATLAS to write annotations to an external file instead.

Default keyboard shortcuts

Once the GUI is running, the following keys control playback and annotation out of the box:

Key

Action

S

Start / stop annotating an action

Q / E

Load previous / next segment

A / D

Step backwards / forwards by ff_value_big

Z / C

Step backwards / forwards by ff_value_small

F

Jump to the end of the current segment

Space

Play / pause

Backspace

Delete the most recent annotation

P

Save annotations for the current segment

All shortcuts are configurable — see Configuration Guide for details.

Next steps

  • Configuration Guide — full reference for every configuration key, with per-format examples.

  • atlas_gui — auto-generated API reference for developers extending the tool with new dataset types.