Hugging Face's logo

Datasets:
garrying
/
OpenScan

Tasks:
Text Classification
Image Segmentation
Modalities:
Text
Formats:
text
Languages:
English
Size:
100M - 1B
ArXiv:
Tags:
3d-scene-understanding
open-vocabulary
scannet200
benchmark
aaaai-2026
Libraries:
Datasets
License:
Dataset card Data Studio Files
xet
 Community
1
OpenScan
File size: 2,506 Bytes
6b6ee6b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
024ce12
6b6ee6b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
024ce12
6b6ee6b
 
024ce12
6b6ee6b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
---
license: other
language:
- en
task_categories:
- text-classification
- image-segmentation
pretty_name: OpenScan
size_categories:
- 10K<n<100K
tags:
- 3d-scene-understanding
- open-vocabulary
- scannet200
- benchmark
- aaaai-2026
---

# OpenScan

OpenScan is a benchmark for generalized open-vocabulary 3D scene understanding. It extends ScanNet200-style scene understanding with eight object-attribute aspects: material, affordance, property, type, manner, synonyms, requirement, and element.

This dataset mirror contains the OpenScan validation annotations and evaluation ground-truth files for the AAAI 2026 paper.

## Files

| File | Description |
| --- | --- |
| `openscan_val_raw.json` | Raw validation annotations with one record per scene object. |
| `openscan_label_mapping/` | Expanded TSV label mappings for all eight OpenScan aspects. |
| `openscan_val_gt/` | Processed validation ground-truth files for evaluation. |

The raw validation JSON contains 10,388 annotated objects from 312 validation scenes. The processed ground truth contains 312 scene files for each of the eight OpenScan aspects.

## Annotation Format

Each raw annotation record has this structure:

```json
{
  "scene_id": "scene0011_00",
  "object_id": "0",
  "object_name": "floor",
  "material": "stone",
  "affordance": "stand",
  "property": "other",
  "type": "land",
  "manner": "other",
  "synonyms": "other",
  "requirement": "other",
  "element": "other"
}
```

## Usage

Load the raw validation annotation table with `datasets`:

```python
from datasets import load_dataset

dataset = load_dataset(
    "json",
    data_files="hf://datasets/garrying/OpenScan/openscan_val_raw.json",
    split="train",
)
```

Download the processed validation ground truth for evaluation:

```bash
hf download garrying/OpenScan --repo-type dataset --local-dir . --include "openscan_val_gt/**"
```

The `openscan_val_gt/` directory contains:

```text
gt_affordance/
gt_element/
gt_manner/
gt_material/
gt_property/
gt_requirement/
gt_synonyms/
gt_type/
```

## Project

- Project page: https://youjunzhao.github.io/OpenScan/
- Code and evaluation: https://github.com/YoujunZhao/OpenScan
- Paper: https://arxiv.org/abs/2408.11030

## Citation

```bibtex
@article{zhao2024openscan,
  title={OpenScan: A Benchmark for Generalized Open-Vocabulary 3D Scene Understanding},
  author={Zhao, Youjun and Lin, Jiaying and Ye, Shuquan and Pang, Qianshi and Lau, Rynson WH},
  journal={arXiv preprint arXiv:2408.11030},
  year={2024}
}
```