Datasets:

Yong-Hoon
/

smplx_anything

license: other
task_categories:
  - other
tags:
  - smplx
  - 3d-human
  - pose-estimation
  - synthetic
pretty_name: SMPLX Anything
size_categories:
  - 1M<n<10M
viewer: false

SMPLX Anything

smplx_anything is a unified-format preprocessed bundle (processed_*_resized) of four SMPL-X-based 3D human datasets, packaged into a single tar archive for easy distribution.

Sub-datasets

Sub-dataset	Folder	Project page	Paper
AGORA	`processed_agora_resized/`	agora.is.tue.mpg.de	Patel et al., CVPR 2021
Anny-One	`processed_annyone_resized/`	Anny-One @ NAVER LABS Europe · GitHub	Baradel et al., 2025
BEDLAM	`processed_bedlam_resized/`	bedlam.is.tue.mpg.de	Black et al., CVPR 2023
BEDLAM 2.0	`processed_bedlam2_resized/`	bedlam2.is.tue.mpg.de	BEDLAM 2.0, NeurIPS 2025

Each sub-dataset retains the license of its original authors. Please review and comply with the original licenses before use.

File layout

The full bundle is packaged into a single tar archive and split into 40 GB chunks for upload.

smplx_anything.tar.aa
smplx_anything.tar.ab
smplx_anything.tar.ac
...
SHA256SUMS         # integrity checksums for each split part
README.md

The split-file suffixes (aa, ab, ...) follow GNU split's default scheme.

Download

1) Using the `hf` CLI (recommended)

pip install -U "huggingface_hub>=1.0"

hf download Yong-Hoon/smplx_anything \
    --repo-type dataset \
    --local-dir ./smplx_anything

The legacy huggingface-cli was deprecated in v1.0 and replaced by hf. If you must use the old CLI: huggingface-cli download Yong-Hoon/smplx_anything --repo-type dataset --local-dir ./smplx_anything

2) Using `git lfs`

git lfs install
git clone https://huggingface.co/datasets/Yong-Hoon/smplx_anything

3) Downloading only some parts

hf download Yong-Hoon/smplx_anything \
    smplx_anything.tar.aa smplx_anything.tar.ab \
    --repo-type dataset \
    --local-dir ./smplx_anything

Integrity check (optional)

cd ./smplx_anything
sha256sum -c SHA256SUMS

Extraction

Concatenate the split parts and pipe them straight into tar. You do not need to first reassemble a single .tar file on disk.

cd ./smplx_anything
cat smplx_anything.tar.* | tar -xvf -

After extraction, the following four folders will appear:

processed_agora_resized/
processed_annyone_resized/
processed_bedlam_resized/
processed_bedlam2_resized/

If you are tight on disk space, you can delete the split parts after extraction. However, in case extraction fails midway, we recommend running sha256sum -c SHA256SUMS first and only deleting the parts after a clean extraction.

If you prefer a single `.tar` file

cat smplx_anything.tar.* > smplx_anything.tar
tar -xvf smplx_anything.tar

Windows users

Concatenate the split parts in PowerShell, then extract with 7-Zip or WinRAR.

Get-Content .\smplx_anything.tar.* -Raw -Encoding Byte | Set-Content .\smplx_anything.tar -Encoding Byte
# Then extract smplx_anything.tar with 7-Zip

How the splits were produced (reproducibility)

The uploaded split files were created with the commands below.

# Bundle the four folders into one tar stream and split into 40 GB chunks
tar -cf - \
    processed_agora_resized \
    processed_annyone_resized \
    processed_bedlam2_resized \
    processed_bedlam_resized \
  | split -b 40G - smplx_anything.tar.

# Generate integrity checksums
sha256sum smplx_anything.tar.* > SHA256SUMS

tar -cf - — bundle the four folders into a tar stream on stdout (no compression).
split -b 40G - — read stdin and split it into 40 GB chunks; suffixes default to aa, ab, ...
Output file prefix: smplx_anything.tar.

No additional compression is applied at the tar level: the underlying media is already compressed (images, etc.), so further compression yields little gain and slows down extraction.

Citation

Please cite the original paper for each sub-dataset you use.

AGORA: Patel et al., AGORA: Avatars in Geography Optimized for Regression Analysis, CVPR 2021. — Project page
Anny-One / Anny: Baradel et al., Human Mesh Modeling for Anny Body, 2025. — Project page · Code
BEDLAM: Black et al., BEDLAM: A Synthetic Dataset of Bodies Exhibiting Detailed Lifelike Animated Motion, CVPR 2023. — Project page
BEDLAM 2.0: BEDLAM 2.0: Synthetic Humans and Cameras in Motion, NeurIPS 2025 (Datasets & Benchmarks). — Project page

Contact

For dataset-related issues, please use the Discussions tab on this repository.