SLAM-MER

A modern visual SLAM framework for map creation, localization, and rapid prototyping. It is a C++ visual SLAM library organized around modular mapping, localization, adjustment, and visualization components. It is designed to be flexible, with optional modules for depth inference and visual place localization.

Note: This codebase is still under active development. If you find any issues or would like to discuss anything, please open an issue or contact Pedro Miraldo.

Usage

Download

Clone the repository and initialize submodules:

# Clone SLAM-MER
git clone https://github.com/slam-mer/slam-mer
# Go to the repository directory
cd slam-mer
# Sync optional module URLs (auto-selects private repos when this repository
# name ends with "_private")
./scripts/sync_optional_modules.sh
# Initialize and update submodules
git submodule update --init --recursive

Compilation

For simplicity, we handle installation and configuration for this repository with mise. For detailed instructions, check the online documentation.

The code has been tested on Ubuntu 24.04 with CUDA 12.8, CUDA_ARCH=86, and TORCH_CUDA_ARCH_LIST=8.6. To simplify compilation, we use mise tasks. Please make sure mise is installed on your machine. Check and update mise.toml accordingly, especially the CUDA version and architecture. Then run:

# Set up trusted configuration
mise trust
# Install tools
mise install
# Create local Python environment
uv sync

Optional Python extras:

# Development tooling (pre-commit)
uv sync --extra dev
# Documentation tooling (Sphinx, Breathe, Exhale)
uv sync --extra docs

To install and register pre-commit hooks:

mise run setup_precommit

It is also recommended to install and configure GitHub CLI for downloading ONNX models from GitHub Releases.

Run the following commands from the root of slam-mer to build from scratch:

# Install system dependencies
mise run install_dependencies
# Build third-party dependencies
mise run build_thirdparties
# Configure SLAM-MER extensions
mise run configure_modules
# Build extensions
mise run build_modules
# Build SLAM-MER libraries
mise run build # or: mise run debug

To compile the core project without building or linking the depth inference and visual place localization modules, skip mise run configure_modules and mise run build_modules, then run:

mise run build_thirdparties
mise run build_without_depth_vpl

For the complete list of configurable CMake build flags, see Compilation Options.

To control which transformation templates are compiled, use -DSLAM_TRANSFORMATION_INSTANTIATIONS=ALL|POSE3|SIMILARITY3 (default: ALL).

What each step does:

1. ./scripts/sync_optional_modules.sh Sets optional module submodule URLs based on the current repository naming: repositories ending with _private use private optional module repos.

2. git submodule update --init --recursive Initializes third-party submodules used by the project, including thirdparty/gtsam, thirdparty/PoseLib, thirdparty/faiss, thirdparty/depth_inference, and thirdparty/visual_place_localization.

3. mise run install_dependencies Installs Ubuntu packages used by the standard build, including Ninja, GCC/G++, CUDA Toolkit 12.8, Boost, GTSAM prerequisites, OpenBLAS/LAPACK, YAML-CPP, Eigen3, TBB, libpng, libjpeg-turbo, and cuDNN when available.

4. mise run build_thirdparties Builds and installs core dependencies required by SLAM-MER:

   • GTSAM (thirdparty/gtsam/build/install)
   • Rerun SDK (thirdparty/rerun_cpp_sdk/build/install)
   • PoseLib (thirdparty/PoseLib/build/install)
   • FAISS with GPU support (thirdparty/faiss/build/install)

5. mise run configure_modules Prepares the module repositories used by SLAM-MER by running their local mise trust, mise install, and uv sync workflows:

   • thirdparty/depth_inference
   • thirdparty/visual_place_localization

6. mise run build_modules Builds the two optional modules used by the default mise run build profile:

   • depth_inference: downloads ONNX Runtime and depth-model assets, then builds and installs to thirdparty/depth_inference/build/install
   • visual_place_localization: downloads ONNX Runtime and MegaLoc assets, builds FAISS for this module, then builds and installs to thirdparty/visual_place_localization/build/install

7. mise run build Downloads thirdparty/libtorch when missing, then configures and builds SLAM-MER libraries with CMake + Ninja and installs them to build/install. This task enables ALIKED, depth inference, and visual place localization by default.

Examples

Examples are maintained in a separate project:

• https://github.com/slam-mer/slam-mer_examples

Please follow the instructions in that repository to build and run examples. If you find any issues, please let us know.

Documentation

We also provide a Sphinx website with useful information about the code structure, class documentation, and contribution workflow. Use the following instructions to generate the documentation:

mise run install_docs_dependencies
mise run build_docs

If you prefer managing Python docs packages with uv, install the docs extra before building (with doxygen available on your system):

uv sync --extra docs
mise run build_docs

Citation

If you use SLAM-MER in your research, please cite:

@inproceedings{merl2026revisiting,
  author = {Piedade, Valter and Manam, Lalit and Yamazaki, Masashi and Miraldo, Pedro},
  title = {Revisiting Monocular SLAM with Spatio-Temporal Scene Modeling},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year = {2026}
}

Contact

Pedro Miraldo Senior Principal Research Scientist Mitsubishi Electric Research Laboratories miraldo@merl.com

Valter Piedade PhD Student in Electrical and Computer Engineering Instituto Superior Tecnico, Lisbon (MERL consultant)

Masashi Yamazaki Advanced Technology Center Mitsubishi Electric Corporation

Contributing

See CONTRIBUTING.md for our policy on contributions.

License

Released under the AGPL-3.0-or-later license, as described in LICENSE.md.

All files, except those listed in LICENSE-THIRD-PARTY.md, have:

Copyright (C) 2026 Mitsubishi Electric Research Laboratories (MERL).

SPDX-License-Identifier: AGPL-3.0-or-later