TripVVT-10K Dataset

News

• 2026.06: TripVVT has been accepted by ECCV 2026.
• 2026.04: The TripVVT paper is available on arXiv.
• The project page is available at https://shaodingbao.github.io/TripVVT/.

TripVVT-10K is a large-scale dataset for in-the-wild Video Virtual Try-On (VVT). It contains 10,031 high-quality video samples with triplet supervision, covering upper-body garments, lower-body garments, and dresses.

TripVVT-10K is released together with the TripVVT paper: TripVVT: A Large-Scale Triplet Dataset and a Coarse-Mask Baseline for In-the-Wild Video Virtual Try-On.

Links

• Dataset: Hugging Face / TripVVT-10K
• Paper: arXiv:2604.27958
• Project page: TripVVT Homepage

Dataset Statistics

Item	Value
Total samples	10,031
Train samples	9,931
Test samples	100
Video resolution	720 x 1280
Garment categories	upper_body / lower_body / dresses
Data formats	MP4 videos, PNG images
Sharded release	18 tar shards, about 72.6 GiB source data

Directory Structure

TripVVT-10K/
+-- data_shards/              # Tar shards for stable download
+-- input/                    # Input videos after extraction
+-- groundtruth/              # Ground-truth target videos after extraction
+-- input_cloth_ch/           # Garment images after extraction
+-- input_cloth_ch_line/      # Garment line images after extraction
+-- video_input_mask/         # Video masks after extraction
+-- pose_dw/                  # Pose videos after extraction
+-- metadata.json             # Full metadata
+-- train.json                # Training split metadata
+-- test.json                 # Test split metadata
+-- train.jsonl               # Training split, one sample per line
+-- test.jsonl                # Test split, one sample per line

Metadata Format

Each sample in metadata.json, train.json, test.json, train.jsonl, and test.jsonl contains fields such as:

{
  "output": "groundtruth/00001.mp4",
  "cloth_type": "upper_body",
  "mask_condition": "video_input_mask/00001.mp4",
  "input": "input/00001_00001_wearing_007324_1.mp4",
  "pose_dw": "pose_dw/00001.mp4",
  "text": "Detailed scene and garment description...",
  "height": 1280,
  "width": 720,
  "line_image": "input_cloth_ch_line/00001_upper_body.png",
  "cloth_image": "input_cloth_ch/00001_upper_body.png",
  "type": "video"
}

Usage

Load the metadata splits from Hugging Face:

from datasets import load_dataset

data_files = {
    "train": "hf://datasets/TripVVT/TripVVT-10K/train.jsonl",
    "test": "hf://datasets/TripVVT/TripVVT-10K/test.jsonl",
}
dataset = load_dataset("json", data_files=data_files)

The loaded rows contain relative paths to the video and image files in the repository.

For a stable full-data download, we recommend downloading the tar shards instead of thousands of individual files:

from huggingface_hub import snapshot_download

local_dir = snapshot_download(
    repo_id="TripVVT/TripVVT-10K",
    repo_type="dataset",
    allow_patterns=[
        "README.md",
        "LICENSE",
        "metadata.json",
        "train.json",
        "test.json",
        "train.jsonl",
        "test.jsonl",
        "data_shards/*",
    ],
)

Extract the shards from the downloaded folder:

cd /path/to/downloaded/TripVVT-10K
for f in data_shards/*.tar; do tar -xf "$f"; done

After extraction, the original folder structure (input/, groundtruth/, input_cloth_ch/, etc.) is restored, and the relative paths in the metadata files can be used directly.

Load metadata locally:

import json

with open("train.json", "r", encoding="utf-8") as f:
    train_metadata = json.load(f)

for sample in train_metadata["pairs"]:
    print(sample["input"])
    print(sample["cloth_image"])
    print(sample["text"])

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license.

Citation

If you use TripVVT-10K, TripVVT-Bench, or the TripVVT project in your research, please cite:

@article{shao2026tripvvt,
  title={TripVVT: A Large-Scale Triplet Dataset and a Coarse-Mask Baseline for In-the-Wild Video Virtual Try-On},
  author={Shao, Dingbao and Wu, Song and Wang, Shenyi and Wang, Ye and Tang, Ziheng and Liu, Fei and Lin, Jiang and Chen, Xinyu and Wang, Qian and Tai, Ying and Yang, Jian and Yi, Zili},
  journal={arXiv preprint arXiv:2604.27958},
  year={2026}
}