Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,86 @@
|
|
1 |
-
---
|
2 |
-
license: cc-by-sa-4.0
|
3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: cc-by-sa-4.0
|
3 |
+
---
|
4 |
+
# TransFrag27K: Transparent Fragment Dataset
|
5 |
+
|
6 |
+
## Dataset Summary
|
7 |
+
This script was used to create the **first large-scale transparent fragment dataset, TransFrag27K**, which contains **27,000 images and masks** at a resolution of **640×480**. The dataset covers fragments of common everyday glassware and incorporates **more than 150 background textures** and **100 HDRI environment lightings**.
|
8 |
+
|
9 |
+
Transparent objects, being a special category, have **refractive and transmissive material properties** that make their visual features highly sensitive to environmental lighting and background. In real-world scenarios, collecting data of transparent objects with diverse backgrounds and lighting conditions is challenging, and annotations are prone to errors due to difficulties in recognition.
|
10 |
+
|
11 |
+
To address this, we designed an **automated dataset generation pipeline in Blender**:
|
12 |
+
- Objects are randomly fractured using the **Cell Fracture plugin**.
|
13 |
+
- Parametric scripts batch-adjust lighting, backgrounds, and camera poses.
|
14 |
+
- Rendering is performed automatically to output paired RGB images and binary masks.
|
15 |
+
|
16 |
+
The dataset is generated using this Blender script. Please refer to:
|
17 |
+
[GitHub Repository](https://github.com/Keithllin/Transparent-Fragments-Contour-Estimation)
|
18 |
+
|
19 |
+
---
|
20 |
+
|
21 |
+
## Related Paper
|
22 |
+
**Transparent Fragments Contour Estimation via Visual-Tactile Fusion for Autonomous Reassembly**
|
23 |
+
*(Lin et al., 2025)*
|
24 |
+
|
25 |
+
---
|
26 |
+
|
27 |
+
## Supported Tasks
|
28 |
+
- **Semantic Segmentation**
|
29 |
+
|
30 |
+
---
|
31 |
+
|
32 |
+
## Dataset Structure
|
33 |
+
In our released dataset, to facilitate subsequent customized processing, we organize each object’s data in the following structure:
|
34 |
+
|
35 |
+
```├─TransFrag27K
|
36 |
+
│ ├─Planar1
|
37 |
+
│ │ ├─anno_mask
|
38 |
+
│ │ └─rgb
|
39 |
+
│ ├─Planar2
|
40 |
+
│ │ ├─anno_mask
|
41 |
+
│ │ └─rgb
|
42 |
+
│ ├─Curved1
|
43 |
+
│ │ ├─anno_mask
|
44 |
+
│ │ └─rgb
|
45 |
+
│ ├─Curved2
|
46 |
+
│ │ ├─anno_mask
|
47 |
+
│ │ └─rgb
|
48 |
+
│ ├─Irregular1
|
49 |
+
│ │ ├─anno_mask
|
50 |
+
│ │ └─rgb
|
51 |
+
│ ├─Irregular2
|
52 |
+
│ │ ├─anno_mask
|
53 |
+
│ │ └─rgb
|
54 |
+
│ ├─Irregular3
|
55 |
+
│ │ ├─anno_mask
|
56 |
+
│ │ └─rgb
|
57 |
+
```
|
58 |
+
|
59 |
+
|
60 |
+
We mainly organize the dataset according to the **shape classes** of transparent fragments:
|
61 |
+
- **Planar**
|
62 |
+
Mainly includes fragments from flat regions such as dish bottoms and glass bases.
|
63 |
+
- **Curved**
|
64 |
+
Mainly includes fragments from objects with cylindrical or spherical curvature, such as cups, bottles, and bowls.
|
65 |
+
- **Irregular**
|
66 |
+
Mainly includes fragments with multiple curvature patterns or discontinuous surfaces, such as the intersection of a cup wall and bottom, special bottle necks, wine glass stems, and handles.
|
67 |
+
|
68 |
+
---
|
69 |
+
|
70 |
+
## Usage Example
|
71 |
+
```python
|
72 |
+
from datasets import load_dataset
|
73 |
+
|
74 |
+
dataset = load_dataset("your-username/TransFrag27K")
|
75 |
+
print(dataset["train"][0])
|
76 |
+
```
|
77 |
+
##Citation
|
78 |
+
If you use this dataset, please cite the following paper:
|
79 |
+
```
|
80 |
+
@article{lin2025transparent,
|
81 |
+
title={Transparent Fragments Contour Estimation via Visual-Tactile Fusion for Autonomous Reassembly},
|
82 |
+
author={Lin, XXX and others},
|
83 |
+
year={2025},
|
84 |
+
journal={TBD}
|
85 |
+
}
|
86 |
+
```
|