Hugging Face's logo

Datasets:
polygom-team
/
2K2K

Tasks:
Image-to-3D
Languages:
English
Tags:
3D human reconstruction
human
SMPLX
SMPL
2K2K
dataset
License:
Dataset card Files Community
2K2K / README.md
polygom-team's picture
polygom-team
Update README.md
3604726 verified
preview code
|
raw
history blame contribute delete
6.39 kB
metadata
viewer: false
license: other
annotations_creators:
  - expert-generated
language:
  - en
task_categories:
  - image-to-3d
tags:
  - 3D human reconstruction
  - human
  - SMPLX
  - SMPL
  - 2K2K
  - dataset
  - cvpr2023
  - 3D human digitization
extra_gated_prompt: >-
  The dataset is encrypted to prevent unauthorized access. Please fill out the
  request form : https://forms.gle/oiTkbADGALga9XP18
extra_gated_fields:
  Name: text
  Academic Institution: text
  E-Mail: text
  Main purpose of data access:
    type: select
    options:
      - Research
      - Education
      - label: Other
        value: other
  I have signed the GOOGLE form before this request: checkbox
pretty_name: >-
  CVPR2023 - High-fidelity 3D Human Digitization from Single 2K Resolution
  Images (2K2K)

NOTE

Please fill out the google form before requesting data access in Huggingface. (with valid "Agreement" PDF file.)

If you skip it, we will ignore/reject the data access.

-> Google form link

Updated (20250414, [email protected])

I'm sorry for mis-typo in the file names and unkind explanation.

I've updated the names and attached simple sample loader based on Open3D.

If you have another issue, please contact via above e-mail.

Polygom2K2K Dataset

High-fidelity 3D Human Digitization from Single 2K Resolution Images

Sang-Hun Han, Min-Gyu Park, Ju Hong Yoon, Ju-Mi Kang, Young-Jae Park and Hae-Gon Jeon. CVPR 2023

[Project Page]

  • This repository provides new 2K2K dataset updated by Polygom.
  • Specifically, we provides same 1M mesehs of original 2K2K dataset with newly updated SMPLX parameters and registered SMPLX via non-rigid fitting.
  • Both SMPLX rigid and non-rigid fittings are simply optimized by noisy 3D keypoint and chamfer distance to the nearest point. Therefore noticable misalignment can be observed in some region (e.g. hair, hand, and foot). If you need accurately aligned SMPLX, please postprocess it.

Prerequisites

Please downalod "SMPLX_NEUTRAL.pkl" and "J_regressor_body25_smplx.txt" from SMPLX official repo

Then, place them into same directory 

$MODEL_ROOT
  |-SMPLX_NEUTRAL.PKL
  |-J_regressor_body25_smplx.txt

$DATA_ROOT
  |- train
  |   |-1M
  |- test

Data explanation

Polygom2K2K includes 2,050(train 2000, test 50) high-fidility 3D body scans captured by 80 multi-view DSLR cameras by POLYGOM. Each scan is simplified to a maximum of 1M vertices. Scans with fewer than 1M vertices are provided in their raw form. The data is provided in PLY format, including vertex colors.

Additionally, we provide the results of SMPLX rigid fitting and SMPL (subdivided into 27554 vertices) non-rigid registration.

  1. Raw meshes are normalized in [0, 1] x [0, 1] x [0, 1] unit cube and centerd at (x0.5, y0.0, z0.5). (ground plane is xz plane and up-direction is +y.)

  2. SMPLX parameters from rigid fitting consist of "poses", "shapes", "expression", "Rh" (global orientation), and "Th" (global translation). Reconstruct it into mesh using the SMPLX layer then multiply "scale".

  3. SMPLX final meshes from non-rigid fitting are optimized by pytorch-nicp

Data access

  1. Please check this link and fill our google form.

  2. After submitting the form, request above data access.

  3. We will double-check the form and the request and approve in 2~3 working days.

Preview data

We've uploaded sample loading code named "load_sample.py".

# pip install open3d required

python load_sample.py --data_root DATA_ROOT --model_root MODEL_ROOT

# e.g. python load_sample.py --data_root /media/ssd/2k2k --model_root /media/smplx

It will visualize overall sample one-by-one using Open3D visualizer.

Agreement

  1. The 2K2K dataset (the "Dataset") is available for non-commercial research purposes only. Any other use, in particular any use for commercial purposes, is prohibited. This includes, without limitation, incorporation in a commercial product, use in a commercial service, as training data for a commercial product, for commercial ergonomic analysis (e.g. product design, architectural design, etc.), or production of other artifacts for commercial purposes including, for example, web services, movies, television programs, mobile applications, or video games. The Dataset may not be used for pornographic purposes or to generate pornographic material whether commercial or not. The Dataset may not be reproduced, modified and/or made available in any form to any third party without POLYGOM’s prior written permission.

  2. You agree not to reproduce, modify, duplicate, copy, sell, trade, resell, or exploit any portion of the images and any portion of derived data in any form to any third party without POLYGOM’s prior written permission.

  3. You agree not to further copy, publish, or distribute any portion of the Dataset. Except, for internal use at a single site within the same organization it is allowed to make copies of the Dataset.

  4. POLYGOM reserves the right to terminate your access to the Dataset at any time.

Citation

If you use this dataset for your research, please consider citing:

@InProceedings{han2023Recon2K,
title={High-fidelity 3D Human Digitization from Single 2K Resolution Images},
author={Han, Sang-Hun and Park, Min-Gyu and Yoon, Ju Hong and Kang, Ju-Mi and Park, Young-Jae and Jeon, Hae-Gon},
booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR2023)},
month={June},
year={2023},
}

@Misc{easymocap,  
    title = {EasyMoCap - Make human motion capture easier.},
    howpublished = {Github},  
    year = {2021},
    url = {https://github.com/zju3dv/EasyMocap}
}