MichaelGrupp/evo

Python package for the evaluation of odometry and SLAM

evo – Evaluation tools for odometry & SLAM (Python)

What it is

  • A Python package that provides command‑line tools and a small library for loading, comparing, and visualising trajectories produced by odometry or SLAM systems.
  • Works on Linux, macOS, Windows and can read data directly from ROS 1/ROS 2 bag files.

Why you might need it

  • Different SLAM/odometry pipelines output trajectories in many formats (TUM, KITTI, EuRoC, ROS topics, bag files). evo normalises these formats so you can evaluate them side‑by‑side.
  • Implements common error metrics such as Absolute Pose Error (APE) and Relative Pose Error (RPE) with options for alignment, scale correction (useful for monocular SLAM), and trajectory association.
  • Generates publication‑ready plots (2‑D, 3‑D, map‑tiled, LaTeX‑compatible) and can export results as CSV/Excel tables.
  • Optional integrations:
    • PyQt6 for a richer GUI backend.
    • contextily for adding geographic map tiles.
    • rerun‑sdk for streaming results to the Rerun visualiser.

Key components

Tool Purpose
evo_ape Compute absolute pose error between a reference trajectory and one or more estimates.
evo_rpe Compute relative pose error (drift over a fixed distance/angle).
evo_traj Plot, analyse, or export trajectories; supports many output formats and map overlays.
evo_res Aggregate multiple metric result files, produce summary statistics and comparative plots.
evo_config Global configuration (e.g., plot backend).
evo_ipython Launch an IPython shell pre‑loaded with the library for interactive exploration.

Installation

# Recommended: install into an isolated environment (venv, uv, pipx, etc.)
pip install evo          # from PyPI – gets the latest stable release
# or, for development:
git clone https://github.com/MichaelGrupp/evo.git
cd evo
pip install --editable .
  • Requires Python 3.10+.
  • Optional dependencies (PyQt6, contextily, rerun‑sdk, ROS) are installed automatically if you request them, but the core functionality works without them.

Typical workflow (example)

  1. Plot trajectories
    cd test/data
evo_traj kitti KITTI_00_ORB.txt KITTI_00_SPTAM.txt --ref=KITTI_00_gt.txt -p --plot_mode=xz
    Produces side‑by‑side plots of the two SLAM runs against ground truth.
  2. Run a metric
    mkdir results
evo_ape kitti KITTI_00_gt.txt KITTI_00_ORB.txt -va --plot --plot_mode xz \
        --save_results results/ORB.zip
    Calculates APE, shows a plot, and stores the detailed output for later aggregation.
  3. Compare multiple results
    evo_res results/*.zip -p --save_table results/table.csv
    Generates summary statistics, violin/box plots, and a CSV table.

Extensibility

  • The package is split into a core library (metric implementations, trajectory handling) and tools (CLI wrappers). You can import the library in your own Python code to build custom evaluation pipelines or integrate with other research frameworks.

Documentation & help

  • Full CLI reference via --help on each command.
  • A detailed wiki on GitHub covers supported formats, plotting options, and Rerun integration.
  • Jupyter notebook metrics_tutorial.ipynb demonstrates interactive use.

License

  • GPL‑3.0‑or‑later. If you use evo in academic work, the README asks for a citation (BibTeX provided).

In short, evo is a mature, well‑tested toolbox that lets robotics researchers quickly benchmark odometry/SLAM outputs without writing custom parsers or plotting code.

Related

  • Project
    MINT-SJTU/Evo-1

    Evo-1 is a lightweight Vision-Language-Action (VLA) model for robotic manipulation that translates visual and textual instructions into robot actions across various simulations and real hardware.

  • Project
    PRBonn/rko_lio

    RKO‑LIO is a sensor‑agnostic LiDAR‑Inertial odometry library that fuses LiDAR scans and IMU data to estimate robot pose. It works on ROS bags (ROS 1/2) via a simple Python CLI (`pip install rko_lio`) or as a ROS 2 package (`apt install ros-<distro>-rko-lio`). Only the IMU‑to‑base and LiDAR‑to‑base extrinsics are required; no per‑sensor modeling is needed. The system includes optional real‑time visualization with the Rerun SDK and is backed by a peer‑reviewed IEEE RA‑L paper.

  • Project
    7df-lab/devo

    An open-source, model-neutral coding agent with a Rust runtime that supports private environments and multiple LLM providers via a Desktop app or TUI/CLI.

  • Project
    Tau-J/rtmlib

    rtmlib is a lightweight Python library for human and animal pose estimation that wraps RTMPose and ViTPose models. It avoids heavy OpenMMLab dependencies, requiring only numpy, OpenCV and ONNX Runtime (with optional GPU/accelerator back‑ends). The package provides high‑level solution classes (Wholebody, Body, Hand, Animal, Custom, Wholebody3d) that combine a detector (YOLOX, RFDETR, RTMDet) and a pose estimator, plus utilities for drawing skeletons. Models are auto‑downloaded from OpenMMLab or a HuggingFace mirror. Install via pip or from source, and run a Gradio WebUI or use the Python API for single‑image, video, or webcam inference.

  • Project
    Tencent/YOLO-Master

    A YOLO-style framework for real-time object detection that uses Mixture-of-Experts (MoE) to adaptively allocate computation based on scene complexity for higher precision and lower latency.