---
title: UHR-BAT-SFT-10K
canonical_url: "https://www.modelscope.cn/datasets/KAIWANG/UHR-BAT-SFT-10K"
md_url: "https://www.modelscope.cn/datasets/KAIWANG/UHR-BAT-SFT-10K.md"
repository: KAIWANG/UHR-BAT-SFT-10K
chinese_name: UHR-BAT-SFT-10K
last_updated: 2026-09-16
license: other
storage_size: "17 GB"
downloads: 5435
stars: 0
---

# UHR-BAT-SFT-10K

> UHR-BAT-SFT-10K - KAIWANG 在 ModelScope 开源的数据集。Mirror of RL-MIND/UHR-BAT-SFT-10K; paper links and citations in README.

KAIWANG/UHR-BAT-SFT-10K 是 ModelScope 魔搭社区上的数据集，存储大小 17 GB，采用 other 许可。

- **Repository**: KAIWANG/UHR-BAT-SFT-10K
- **License**: other
- **Storage size**: 17 GB
- **Downloads**: 5435
- **Stars**: 0
- **Last updated**: 2026-09-16

Source: https://www.modelscope.cn/datasets/KAIWANG/UHR-BAT-SFT-10K

---

<table align="center" role="presentation" style="margin:0 auto; border:0; background:transparent;">
  <tr style="border:0; background:transparent;">
    <td align="center" style="border:0; background:transparent; padding:0 8px;">
      <img src="./assets/rl-mind-logo-v2.webp" alt="RL-MIND research group logo" width="160" height="160" loading="eager" fetchpriority="high" decoding="async" style="width:160px; height:160px; object-fit:contain;" />
    </td>
    <td align="center" style="border:0; background:transparent; padding:0 8px;">
      <img src="./assets/dataset-logo-v2.webp" alt="UHR-BAT-SFT-10K dataset logo" width="160" height="160" loading="eager" fetchpriority="high" decoding="async" style="width:160px; height:160px; object-fit:contain;" />
    </td>
  </tr>
</table>

<h1 align="center">UHR-BAT-SFT-10K</h1>
<p align="center"><strong>Supervised Fine-Tuning for Ultra-High-Resolution Remote Sensing</strong></p>

<p align="center">
  <a href="https://arxiv.org/abs/2604.13565"><img src="https://img.shields.io/badge/arXiv-2604.13565-B31B1B.svg" alt="Paper on arXiv" /></a>
  <a href="https://huggingface.co/datasets/RL-MIND/UHR-BAT-SFT-10K"><img src="https://img.shields.io/badge/🤗_Hugging_Face-Dataset-FFD21E" alt="Dataset on Hugging Face" /></a>
  <a href="https://modelscope.cn/datasets/KAIWANG/UHR-BAT-SFT-10K"><img src="https://img.shields.io/badge/ModelScope-Dataset-624AFF" alt="Dataset on ModelScope" /></a>
</p>
<p align="center"><a href="https://yunkaidang.github.io/bibliography/dang2026uhr-bat/">Project</a> · <a href="https://arxiv.org/abs/2604.13565">Paper</a> · <a href="https://github.com/Yunkaidang/UHR-BAT">Code</a></p>
<p align="center"><strong>English</strong> | <a href="./README_ZH.md">中文</a></p>

## 📚 Introduction

**UHR-BAT-SFT-10K** contains visual question answering style instruction-following examples for ultra-high-resolution remote-sensing imagery. It is the supervised fine-tuning dataset used for **UHR-BAT: Budget-Aware Token Compression Vision-Language Model for Ultra-High-Resolution Remote Sensing**.

UHR-BAT allocates visual token budgets according to the current instruction while preserving query-relevant regional evidence. This dataset supports research on efficient multimodal instruction tuning, token compression, and query-aware visual selection.

[Statistics](#statistics) · [Data format](#data-format) · [Quick start](#quick-start) · [Use and notes](#use-and-notes) · [License](#license) · [Citation](#citation)

<a id="statistics"></a>
## 📊 Dataset Statistics

| Item | Value |
|---|---:|
| Training examples | 10,000 |
| Unique image files | 2,422 |
| Image references | 10,089 |
| Single-image examples | 9,911 |
| Multi-image examples | 89 |
| Metadata | 2.46 MiB |
| Packaged image content | 17.17 GiB |

<a id="data-format"></a>
## 📦 Data Format

The release uses an ImageFolder-compatible layout. `train/metadata.parquet` contains one row per SFT example.

```text
train/
├── metadata.parquet
└── images/
```

The metadata field `file_name` points to the first image and is converted by the Hugging Face ImageFolder loader into the `image` feature. Original image references remain in `image_paths`, including both images for multi-image examples.

| Field | Description |
|---|---|
| `image` | First image, exposed by the ImageFolder loader and the Hugging Face Dataset Viewer. |
| `id` | Original example ID. |
| `question` | User question with the leading `<image>` marker removed. |
| `answer` | Assistant answer. |
| `prompt` | Original user prompt, including the `<image>` marker. |
| `conversations` | Original two-turn SFT conversation. |
| `image_paths` | Original relative image path list; paths are resolved under `train/images/`. |
| `num_images` | Number of referenced images. |
| `second_image_path` | Second image path when present; otherwise null. |
| `source_dataset` | Source package name. |
| `first_image_width`, `first_image_height` | First-image dimensions in pixels. |

<a id="quick-start"></a>
## 🚀 Quick Start

Download from ModelScope, then load the local files with the ImageFolder loader in Hugging Face Datasets. The full download includes about 17.17 GiB of images.

```python
from modelscope.hub.snapshot_download import dataset_snapshot_download
from datasets import load_dataset

root = dataset_snapshot_download(
    dataset_id="KAIWANG/UHR-BAT-SFT-10K",
    revision="master",
    local_dir="./UHR-BAT-SFT-10K",
)
dataset = load_dataset("imagefolder", data_dir=f"{root}/train", drop_labels=True)
sample = dataset["train"][0]
```

<a id="use-and-notes"></a>
## 💡 Intended Use and Notes

- Supervised fine-tuning and analysis of multimodal large language models for ultra-high-resolution remote-sensing understanding.
- Research on token compression, query-aware visual selection, regional evidence preservation, and efficient training under limited context budgets.
- The standard `image` feature exposes only the first image. Use `image_paths` to recover all images in a multi-image example.

<a id="license"></a>
## 📜 License and Source

The dataset's existing license declaration is **`other`**. The source card does not specify a license name or license URL; this release preserves that declaration.

Dataset repositories: [Hugging Face](https://huggingface.co/datasets/RL-MIND/UHR-BAT-SFT-10K) · [ModelScope mirror](https://modelscope.cn/datasets/KAIWANG/UHR-BAT-SFT-10K). The `source_dataset` field retains the source package attribution for each example.

<a id="citation"></a>
## 📖 Citation

If you use this dataset or the associated method, please cite:

```bibtex
@article{dang2026uhr,
  title={UHR-BAT: Budget-Aware Token Compression Vision-Language model for Ultra-High-Resolution Remote Sensing},
  author={Dang, Yunkai and Dai, Minxin and Yang, Yuekun and Li, Zhangnan and Li, Wenbin and Miao, Feng and Gao, Yang},
  journal={arXiv preprint arXiv:2604.13565},
  year={2026}
}
```