SIZER Dataset Repository

Download Dataset

https://nextcloud.mpi-klsb.mpg.de/index.php/s/nx6wK6BJFZCTF8C/authenticate/showShare

Fill this form

For dataset access

1. Fill this form
2. For password, send an email to: [email protected]

Installation and Datafiles:

Check INSTALL.MD

Dataset Details

A dataset of clothing size variation of approximately 2000 scans including 100 subjects wearing 10 garment classes in different sizes, where we make available, scans, clothing segmentation, SMPL+G registrations, body shape under clothing, garment class and size labels

Data and Annotations

We compare SIZER dataset with existing real world 3D datasets

Data	Number of Scans	Registrations	Segmentation	Minimal Clothing/Body under clothing	Multiview images	Labels	Demographics
SIZER	~2000	SMPL, SMPL+D, SMPL+G	Upper, Lower and body	yes	code or scanner images on request	clothing style, size and gender	Yes(on request)
CAPE	Dynamic scans	SMPL , SMPL+D	No	Yes	No*	Gender	No
THUman2.0	~500	SMPL , SMPL+X	No	Yes	No*	-	No

Clothing Style	Number of scans
TShirt, Shorts	889
Shirt, Pants	655
Shirt, Shorts	182
Shirt +Coat, Pants	252
Hoodies, Pants	255
Vest, Short	226
Vest, Pants	23

Visualize and process data

Visualize scan and clean floor noise

python vis_data/scan_visualise.py --scan=<subjectid>/<scanid> --process remove_floor

This script only visualises original scan and cleaned scan and saves the clean mesh in the same data directory subjectid = {10001, 10005 ....... } scanid = {1937.....} (for 10001)

Output

               

Original Scan         Segmentation Labels         Clean Scan

Visualize segmented garment layers

python vis_data/get_garment.py --scan=<subjectid>/<scanid>

Output

               

Original Scan         Segmentation Labels         Clean Scan

This script only visualises original scan and 3 layers of segmented scan, namely upper garment, lower garment and other.

Visualize registration

python vis_data/visualise_registration.py --scan=<subjectid>/<scanid>

Note: Before using/comparing scans and registrations, align scan, using align_scan() in visualise_registration.py

Evaluate registration

python vis_data/visualise_registration.py --scan=<subjectid>/<scanid>

If you have your own code/method for scan registrations, we here provide a code to evaluate the quality of registration.

Others

We here provide code for using/evaluating SIZER dataset for various tasks such as 3D reconstruction from images, scan fitting etc.

Image based reconstruction

For image based reconstruction, SIZER scans can be rendered and data pair of {image, scans, SMPL params} can be generated for training or evaluation.

Rendering using fixed camera views:

python image_recon/pytorch_renderer.py --mesh_path=<obj_file> --out_dir=<out_dir>

Rendering image, depth and normal

python image_recon/image_renderer.py --mesh_path=<obj_file> --out_dir=<out_dir>

<obj_file> should contain <>.obj and <>.jpg in the same folder with same name. Currently we render from 72 fixed views, This can be changed in create_rotmat() function in image_recon/render_utils.py

Output

               

RGB render         Depth         Normal

Creating data pairs:

Coming Soon

If you are using SIZER dataset, please cite:

@inproceedings{tiwari20sizer,
    title = {SIZER: A Dataset and Model for Parsing 3D Clothing and Learning Size Sensitive 3D Clothing},
    author = {Tiwari, Garvita and Bhatnagar, Bharat Lal and Tung, Tony and Pons-Moll, Gerard},
    booktitle = {European Conference on Computer Vision ({ECCV})},
    month = {August},
    organization = {{Springer}},
    year = {2020},
    }

@inproceedings{antic2024close,
    title = {{CloSe}: A {3D} Clothing Segmentation Dataset and Model},
    author = {Antić, Dimitrije and Tiwari, Garvita and Ozcomlekci, Batuhan  and Marin, Riccardo  and Pons-Moll, Gerard},
    booktitle = {International Conference on 3D Vision (3DV)},
    month = {March},
    year = {2024},
}

Also refer to CloSe for diverse poses of SIZER dataset and more accurate and fine-grained clothing segmentaion.