---
title: dex_for_pretrain
canonical_url: "https://www.modelscope.cn/datasets/gerryrex/dex_for_pretrain"
md_url: "https://www.modelscope.cn/datasets/gerryrex/dex_for_pretrain.md"
repository: gerryrex/dex_for_pretrain
last_updated: 2026-11-01
license: CC-BY-NC-4.0
storage_size: "3.4 GB"
downloads: 8
stars: 0
---

# dex_for_pretrain

> dex_for_pretrain - gerryrex 在 ModelScope 开源的数据集。LSA DexYCB Pretraining Data

gerryrex/dex_for_pretrain 是 ModelScope 魔搭社区上的数据集，存储大小 3.4 GB，采用 CC-BY-NC-4.0 许可。

- **Repository**: gerryrex/dex_for_pretrain
- **License**: CC-BY-NC-4.0
- **Storage size**: 3.4 GB
- **Downloads**: 8
- **Stars**: 0
- **Last updated**: 2026-11-01

Source: https://www.modelscope.cn/datasets/gerryrex/dex_for_pretrain

---

# LSA DexYCB Pretraining Data

This repository provides the processed DexYCB data used for pretraining **Lift, See, Act (LSA)**. It contains RGB images, pseudo-depth maps, hand-keypoint heatmaps, and 2D/3D wrist annotations.

## Contents

After extracting `LSA-DexYCB-Pretrain.tar`, the dataset has the following structure:

```text
image/             # RGB images
dp_image/          # pseudo-depth maps
kpmap/             # hand-keypoint heatmaps
hand_keyPoint/     # 3D wrist annotations
hand_keyPoint2D/   # 2D wrist annotations
```

The teacher-feature cache is not included. VGGT and Depth Pro features can be generated offline using the preprocessing scripts in the [LSA code repository](https://github.com/yiyuan00/Lift-See-Act).

## Download and Extract

```bash
modelscope download --dataset gerryrex/dex_for_pretrain --local_dir ./dex_for_pretrain
mkdir -p ./dataset
tar -xf ./dex_for_pretrain/LSA-DexYCB-Pretrain.tar -C ./dataset
```

Optionally verify the archive before extraction:

```bash
cd ./dex_for_pretrain
sha256sum -c SHA256SUMS
```

## License

This dataset is derived from [DexYCB](https://dex-ycb.github.io/), which is licensed under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). It is intended for non-commercial research use only. Users must cite the original DexYCB dataset and comply with its license.

## Citation

Please cite the original DexYCB paper and the LSA paper when using this processed dataset.

## Contact

For questions, please contact 1209027338@qq.com.
