Nymeria and NymeriaPlus Dataset

[Nymeria Data Explorer] [Nymeria Paper] [NymeriaPlus Data Explorer] [NymeriaPlus Paper] [Bibtex]

Nymeria is the world's largest dataset of human motion in the wild, capturing diverse people performing diverse activities across diverse locations. It is first of a kind to record body motion using multiple egocentric multimodal devices, all accurately synchronized and localized in one metric 3D world. Nymeria is also the world's largest motion dataset with natural language descriptions. The dataset is designed to accelerate research in egocentric human motion understanding and presents exciting challenges to advance contextualized computing and future AR/VR technology.

NymeriaPlus is an updated version of the original Nymeria dataset with additional annnotations and data. NymeriaPlus features: (1) optimized human motion in both MHR and SMPL formats; (2) dense 3D and 2D bounding box annotations for indoor objects and structured elements; (3) instance-level 3D object reconstruction based on ShapeR and human rating; and (4) additional modalities such as basemap recordings, wristband videos, headset audio and etc.

⚠️ NOTE (June 11, 2026): We have identified a few sequences with missing data due to an error in our release pipeline. We will patch the affected data as soon as possible — please check back here for updates. We apologize for the inconvenience.

This repository hosts the API for downloading and visualizing the dataset. The main branch best supports the NymeriaPlus dataset. For the Nymeria dataset, switch to the nymeria_dataset_legacy branch if main does not work.

Getting Started

Installation

Clone the repository:

git clone git@github.com:facebookresearch/nymeria_dataset.git
cd nymeria_dataset

Option A: install with uv

This route works well on Linux and partially on MacOS if your workflow does not require loading MHR human motion. The installation requires Python 3.12 and is managed with uv by default. From the repository root:

# Linux/MacOS: install uv if it is not already available
curl -LsSf https://astral.sh/uv/install.sh | sh

uv sync                 # runtime deps, including projectaria-tools, torch, and pymomentum-cpu
uv sync --extra gpu     # use CUDA-supported pymomentum-gpu on supported platforms
uv sync --extra smpl    # add SMPL body model support

pymomentum-cpu depends on PyTorch's shared libraries at runtime. If importing pymomentum fails with a missing libtorch.so, point LD_LIBRARY_PATH at the uv environment's torch library directory before running scripts:

export LD_LIBRARY_PATH="$(pwd)/.venv/lib/python3.12/site-packages/torch/lib:${LD_LIBRARY_PATH:-}"

Option B: install with conda

If your workflow requires loading MHR human motion on MacOS, you need to use Conda to install pymomentum-cpu>=0.1.110 from conda-forge with the matching PyTorch 2.10 runtime, which provides the native dependency stack missing from the experimental PyPI macOS wheel.

conda env create -f environment.yml
conda activate nymeriaplus
python -m pip install -r requirements-conda-pip.txt
python -m pip install -e . --no-deps

Downloading data

Before running the code, you need to obtain a valid URL JSON file for the NymeriaPlus dataset from this site, and for the Nymeria dataset from this site. Both websites allow you to filter sequences according to various attributes, e.g., by location, date, activity scenarios, and available annotations. Nymeria and NymeriaPlus datasets each provide 1100 sequences, which in total amounts to ~80 TB per dataset. For convenience, we group files by groups, as defined in nymeriaplus/layout.py. You can choose which groups to download after selecting the sequences from the corresponding websites.

After obtaining the JSON file, run the following script to download data. For downloading the Nymeria dataset, switch to the nymeria_dataset_legacy branch.

With the uv install:

uv run nymeriaplus-download -i /path/to/url.json -o /path/to/outdir -y

# equivalent invocations
uv run python -m nymeriaplus.cli.download -i /path/to/url.json -o /path/to/outdir -y
uv run python nymeriaplus/cli/download.py -i /path/to/url.json -o /path/to/outdir -y

# or activate the environment first
source .venv/bin/activate
python nymeriaplus/cli/download.py -i /path/to/url.json -o /path/to/outdir -y

With the conda install:

conda activate nymeriaplus
nymeriaplus-download -i /path/to/url.json -o /path/to/outdir -y

The downloader processes every sequence and artifact listed under the JSON sequences field, except video_main_rgb preview videos, which are ignored. Zip artifacts are extracted into each sequence directory. Single-file artifacts are placed according to the JSON sequence_config layout, matching nymeriaplus/layout.py.

Completed artifacts are tracked under <out_dir>/.download_logs/ so interrupted downloads can be resumed without adding extra files to sequence directories. Use --overwrite to redownload artifacts that already have completion markers.

Running the viewer

The nymeriaplus-viewer CLI loads a sequence, synchronizes its streams, and opens an interactive viewer. It supports loading the following modalities:

• RGB video streams of the participant's and the observer's headset
• Device trajectories of the participant's headset/wristbands and the observer's headset
• Semi-dense point cloud from the participant's headset
• Skeleton motion from XSens/MHR/SMPL, skinned mesh from MHR/SMPL
• 3D object bounding boxes and the estimated visible 2DBB projected onto the RGB cameras
• Instance-level object mesh reconstruction

The viewer always tries to load all available modalities found for the input sequence. Missing modalities will be silently skipped. Each loaded modality can be toggled to show/hide via the GUI. The available key bindings are annotated on the GUI, e.g., MHR mesh (m) means the m key toggles MHR rendering, 3D bounding boxes (b) means the b key toggles 3D bounding boxes rendering, etc. Follow the instructions below to run the viewer.

With the uv install:

uv run nymeriaplus-viewer -i /path/to/nymeriaplus/sequence \
  [--smpl-model-path PATH] [--target-fps FLOAT] [--ui-scale FLOAT]

# equivalent invocations
uv run python -m nymeriaplus.cli.viewer -i /path/to/sequence
uv run python nymeriaplus/cli/viewer.py -i /path/to/sequence

# or activate the environment first
source .venv/bin/activate
python nymeriaplus/cli/viewer.py -i /path/to/sequence

With the conda install:

conda activate nymeriaplus
nymeriaplus-viewer -i /path/to/sequence

Viewer options

• -i PATH (required): NymeriaPlus sequence root directory.
• --smpl-model-path PATH: path to a SMPL .pkl model file. Required to visualize the SMPL mesh. To enable this rendering, install the smplx package first via uv sync --extra smpl and download the SMPL model files from https://smpl.is.tue.mpg.de/. NymeriaPlus resolves all human motion using the SMPL neutral model.
• --target-fps FLOAT: synchronization target frame rate (default: 30.0).
• --ui-scale FLOAT: override the ImGui UI scale (range 0.25–4.0). By default the viewer uses the window content scale.

Example with the SMPL mesh enabled (assuming xdata_smpl_neutral.npz is downloaded):

uv run nymeriaplus-viewer \
  -i /data/nymeriaplus/seq_001 \
  --smpl-model-path /models/smpl/SMPL_neutral.pkl \
  --target-fps 30

If pymomentum fails to import with a missing libtorch.so, export LD_LIBRARY_PATH as described in Option A before launching the viewer. If the viewer fails to load a Nymeria sequence, switch to the nymeria_dataset_legacy branch instead.

License

This code is released by Meta under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0). Data and code may not be used for commercial purposes. For more information, please refer to the LICENSE file included in this repository.

Use of the datasets is governed by their own terms, included at the repository root: NYMERIA_DATASET_LICENSE for the Nymeria dataset and NYMERIAPLUS_DATASET_LICENSE for the NymeriaPlus dataset.

Attribution

When using the Nymeria dataset and code, please attribute it as follows:

@inproceedings{nymeria24,
      title={Nymeria: A Massive Collection of Multimodal Egocentric Daily Motion in the Wild},
      author={Lingni Ma and Yuting Ye and Fangzhou Hong and Vladimir Guzov and Yifeng Jiang and Rowan Postyeni and Luis Pesqueira and Alexander Gamino and Vijay Baiyya and Hyo Jin Kim and Kevin Bailey and David Soriano Fosas and C. Karen Liu and Ziwei Liu and Jakob Engel and Renzo De Nardi and Richard Newcombe},
      booktitle={the 18th European Conference on Computer Vision (ECCV)},
      year={2024},
      url={https://arxiv.org/abs/2406.09905},
}

When using the NymeriaPlus dataset, please also attribute the following:

@misc{nymeriaplus26,
      title={NymeriaPlus: Enriching Nymeria Dataset with Additional Annotations and Data},
      author={Daniel DeTone and Federica Bogo and Eric-Tuan Le and Duncan Frost and Julian Straub and Yawar Siddiqui and Yuting Ye and Jakob Engel and Richard Newcombe and Lingni Ma},
      year={2026},
      eprint={2603.18496},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.18496},
}

Contribute

We welcome contributions! Go to CONTRIBUTING and our CODE OF CONDUCT for how to contribute.