thomas-yanxin/multi3drefer-mirror

---
language:
- en
size_categories:
- 10K<n<100K
license: mit
---

# Multi3DRefer Dataset
This repository contains the Multi3DRefer dataset, introduced in [Multi3DRefer: Grounding Text Description to Multiple 3D Objects](https://3dlg-hcvc.github.io/multi3drefer/).

[](https://github.com/3dlg-hcvc/multi3drefer)
[](https://pypi.org/project/torchmetrics-ext/)
[](https://arxiv.org/abs/2309.05251)

## Dataset Attributes
**scene_id: (str)** the scene identifier as defined in [ScanNetv2](http://www.scan-net.org/). \
**object_name: (str)** the name of the target object(s). \
**ann_id: (int)** the local annotation ID within a scene. \
**description: (str)** the language description of the target object(s) in the scene. \
**object_ids: (List[int])** the target object ID(s) as defined in [ScanNetv2](http://www.scan-net.org/). \
**eval_type: (str)** the evaluation type of the data, possible values: "zt" (zero target), "mt" (multiple targets), "st_w_d" (single target with distractors) and "st_wo_d" (single target without distractors). \
**spatial: (bool)** indicates if the description mentions spatial information about the target object(s). \
**color: (bool)** indicates if the description mentions the color of the target object(s). \
**texture: (bool)** indicates if the description mentions texture information about the target object(s). \
**shape: (bool)** indicates if the description mentions shape information about the target object(s).

## Citation
```bibtex
@inproceedings{zhang2023multi3drefer,
author={Zhang, Yiming and Gong, ZeMing and Chang, Angel X},
title={Multi3DRefer: Grounding Text Description to Multiple 3D Objects},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month={October},
year={2023},
pages={15225-15236}
}
```

---
语言：
- 英语
规模类别：
- 10K < n < 100K
许可证：MIT许可证
---

# Multi3DRefer 数据集
本代码仓库包含Multi3DRefer数据集，该数据集由论文《Multi3DRefer: Grounding Text Description to Multiple 3D Objects》提出，论文链接：https://3dlg-hcvc.github.io/multi3drefer/。

[](https://github.com/3dlg-hcvc/multi3drefer)
[](https://pypi.org/project/torchmetrics-ext/)
[](https://arxiv.org/abs/2309.05251)

## 数据集属性
**scene_id：（字符串类型）** 即ScanNetv2（ScanNetv2）中定义的场景标识符。
**object_name：（字符串类型）** 目标对象的名称。
**ann_id：（整数类型）** 单一场景内的本地标注ID。
**description：（字符串类型）** 场景中目标对象的自然语言描述文本。
**object_ids：（整数列表类型）** 即ScanNetv2（ScanNetv2）中定义的目标对象ID列表。
**eval_type：（字符串类型）** 数据的评估类型，可选取值为："zt"（零目标（zero target））、"mt"（多目标（multiple targets））、"st_w_d"（带干扰物的单目标（single target with distractors））以及"st_wo_d"（无干扰物的单目标（single target without distractors））。
**spatial：（布尔类型）** 用于标记描述是否提及目标对象空间信息的属性。
**color：（布尔类型）** 用于标记描述是否提及目标对象颜色信息的属性。
**texture：（布尔类型）** 用于标记描述是否提及目标对象纹理信息的属性。
**shape：（布尔类型）** 用于标记描述是否提及目标对象形状信息的属性。

## 引用格式
bibtex
@inproceedings{zhang2023multi3drefer,
author={Zhang, Yiming and Gong, ZeMing and Chang, Angel X},
title={Multi3DRefer: Grounding Text Description to Multiple 3D Objects},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month={October},
year={2023},
pages={15225-15236}
}