Datasets:

zwpro
/

OmniView

Tasks:

Modalities:

Formats:

Size:

Tags:

Libraries:

License:

Dataset card Data Studio Files Files and versions

xet

Community

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

By requesting access to OmniView, you agree to use the dataset responsibly, comply with the dataset license, and cite BooM-VVT in resulting research.

OmniView

Large-scale multi-view garment–model data for virtual try-on

_{Figure 1. OmniView pairs multi-view model images with front/back flat garments across four categories, and adds pseudo try-on data that prior multi-view sets do not provide.}

OmniView is the multi-view virtual try-on dataset introduced in BooM-VVT: Boosting Mask-Free Video Virtual Try-On with Image-Level Pseudo Data. It provides 8,597 garment–model groups across upper-body, lower-body, outerwear, and full-body categories, totalling 35,904 model images and 15,879 flat-garment images. Notably, it offers 6,873 groups that pair multi-view model images with corresponding front- and back-view flat garments.

✨ What's inside

8,597 curated garment–model groups — 35,904 model images and 15,879 flat-garment images in total
Four garment categories — upper_body, lower_body, outerwear, full_body
6,873 groups provide multi-view model images with corresponding front- and back-view flat garments (labeled M2C2 below)
80.81% of those groups contain at least one back-view model image

📢 Release plan

Release	Status	Contents
Garment–model data	✅ Available	Model images, flat-garment images, viewpoint labels, textual annotations
Pseudo-label data & detailed annotations	🕒 September 2026	Pseudo-labeled data for mask-free try-on training, finer-grained model-image annotations

The BooM-VVT paper reports a 6,110-sample experimental snapshot. Collection, regrouping, annotation refinement, and manual verification continued after that snapshot. The current release contains 6,873 core M2C2 groups, plus the additional M1C1, M1C2, and M2C1 groups produced during curation.

📊 Dataset snapshot

_{Figure 2. Release scale, distribution across configurations and garment categories, and model-view coverage within the M2C2 subset.}

M describes model-image multiplicity and C describes the number of flat-garment views.

Symbol	Definition
`M1`	Exactly one model image
`M2`	At least two model images differing in viewpoint or pose
`C1`	One flat-garment image / view
`C2`	Two flat-garment views: one `front` and one `back`

Configuration	Groups	Model images	Garment images	Avg. models/group	Groups w/ back view
`M1C1`	32	32	32	1.00	2 (6.25%)
`M1C2`	409	409	818	1.00	58 (14.18%)
`M2C1`	1,283	5,825	1,283	4.54	1,224 (95.40%)
`M2C2`	6,873	29,638	13,746	4.31	5,554 (80.81%)
Total	8,597	35,904	15,879	4.18	6,838 (79.54%)

📁 Directory structure

OmniView/
├── metadata.jsonl
├── M1C1/
├── M1C2/
├── M2C1/
└── M2C2/
    └── <category>/                # upper_body | lower_body | outerwear | full_body
        └── <group_id>/             # 000000 ... 008596
            ├── model/
            ├── cloth/
            └── anno.json

metadata.jsonl contains one normalized record per group, ordered by the globally unique group_id. It provides relative image paths, viewpoint labels, garment descriptions, person / outfit / background captions, image counts, and back-view indicators for direct programmatic loading.

🏷️ Annotation format

Each group contains one anno.json file:

{
  "cloth": {
    "views": { "front.jpg": "front", "back.jpg": "back" },
    "major_category": "upper_body",
    "garment_type": "Shirt",
    "short_name": "white cotton shirt",
    "detailed_description": "..."
  },
  "model": [
    {
      "views": { "0.jpg": "front", "1.jpg": "back" },
      "person": "...",
      "outfit": "...",
      "background": "..."
    }
  ],
  "model_detail": {
    "0.jpg": { "model_num": "single" },
    "1.jpg": { "model_num": "single" }
  }
}

Field	Type	Description
`cloth.views`	`dict`	Garment image filename → coarse viewpoint
`cloth.major_category`	`str`	One of the four normalized category names
`cloth.garment_type`	`str`	Fine-grained garment type
`cloth.short_name`	`str`	Concise garment description
`cloth.detailed_description`	`str`	Detailed appearance and construction description
`model[].views`	`dict`	Model image filename → coarse viewpoint
`model[].person`	`str`	Description of the person
`model[].outfit`	`str`	Description of the full outfit in the image
`model[].background`	`str`	Description of the scene / background
`model_detail`	`dict`	Per-image model-count metadata (e.g. `single`)

📖 Citation

If you use OmniView, please cite BooM-VVT:

@inproceedings{zhang2026boomvvt,
  title     = {BooM-VVT: Boosting Mask-Free Video Virtual Try-On with Image-Level Pseudo Data},
  author    = {Zhang, Wei and Li, Xin and Shi, Peishu and Gao, Jialin and
               Peng, Xuekang and Lian, Zhichao and Jin, Yeying},
  booktitle = {Proceedings of the 34th ACM International Conference on Multimedia},
  year      = {2026}
}

⚖️ License

This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

🙏 Acknowledgments

The images in OmniView come from the IGPair dataset and from publicly available sources on the internet. We are grateful to the IGPair authors and to everyone whose work made this data available. All images and brands remain the property of their respective owners.

Downloads last month: 33

Total file size:

15.7 GB