Yong-Hoon/human3r-dataset

Human3R Dataset

Human body reconstruction dataset containing 3DPW, EMDB, RICH, and processed BEDLAM data.

Dataset Overview

Dataset	Size	Files
3DPW	9.7 GB	1 file
EMDB	100 GB	2 parts
RICH	453 GB	10 parts
processed_bedlam	826 GB	17 parts
Total	~1.39 TB	30 files

File Structure

human3r-dataset/
├── 3DPW.tar                      # Single file (9.7 GB)
├── EMDB.tar.part_aa              # 50 GB
├── EMDB.tar.part_ab              # 50 GB
├── RICH.tar.part_aa ~ part_aj    # 10 parts (50 GB each, last part 2.7 GB)
└── processed_bedlam.tar.part_aa ~ part_aq  # 17 parts (50 GB each, last part 26 GB)

Download

Using Hugging Face CLI

# Install huggingface_hub
pip install huggingface_hub

# Download entire dataset
huggingface-cli download Yong-Hoon/human3r-dataset --repo-type dataset --local-dir ./human3r-dataset

# Download specific dataset only
huggingface-cli download Yong-Hoon/human3r-dataset --repo-type dataset --local-dir ./human3r-dataset --include "3DPW.tar"
huggingface-cli download Yong-Hoon/human3r-dataset --repo-type dataset --local-dir ./human3r-dataset --include "EMDB.tar.part_*"
huggingface-cli download Yong-Hoon/human3r-dataset --repo-type dataset --local-dir ./human3r-dataset --include "RICH.tar.part_*"
huggingface-cli download Yong-Hoon/human3r-dataset --repo-type dataset --local-dir ./human3r-dataset --include "processed_bedlam.tar.part_*"

Using Python

from huggingface_hub import snapshot_download

# Download entire dataset
snapshot_download(
    repo_id="Yong-Hoon/human3r-dataset",
    repo_type="dataset",
    local_dir="./human3r-dataset"
)

Merge and Extract

After downloading, merge the split files and extract:

3DPW (No merge needed)

tar -xf 3DPW.tar

EMDB

cat EMDB.tar.part_* > EMDB.tar
tar -xf EMDB.tar
rm EMDB.tar  # Optional: remove after extraction

RICH

cat RICH.tar.part_* > RICH.tar
tar -xf RICH.tar
rm RICH.tar  # Optional: remove after extraction

processed_bedlam

cat processed_bedlam.tar.part_* > processed_bedlam.tar
tar -xf processed_bedlam.tar
rm processed_bedlam.tar  # Optional: remove after extraction

Extract All (One-liner)

# Extract all datasets
tar -xf 3DPW.tar && \
cat EMDB.tar.part_* | tar -xf - && \
cat RICH.tar.part_* | tar -xf - && \
cat processed_bedlam.tar.part_* | tar -xf -

Dataset Structure After Extraction

human3r-dataset/
├── 3DPW/
│   ├── imageFiles/
│   └── sequenceFiles.zip
├── EMDB/
│   └── ...
├── RICH/
│   └── ...
└── processed_bedlam/
    └── ...

Verify Checksums

After downloading, verify file integrity using SHA256 checksums:

# Verify all files
sha256sum -c checksums.sha256

# Verify specific dataset
sha256sum -c checksums.sha256 --ignore-missing 2>/dev/null | grep "3DPW"
sha256sum -c checksums.sha256 --ignore-missing 2>/dev/null | grep "EMDB"
sha256sum -c checksums.sha256 --ignore-missing 2>/dev/null | grep "RICH"
sha256sum -c checksums.sha256 --ignore-missing 2>/dev/null | grep "processed_bedlam"

If all files are intact, you will see OK for each file. If any file shows FAILED, please re-download that file.

License

Please refer to the original dataset licenses:

• 3DPW
• EMDB
• RICH
• BEDLAM