File size: 4,158 Bytes
e4d46d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b0f6f5a
 
 
 
 
 
e4d46d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
538b0cc
e4d46d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
---
license: other
license_name: transparent-460-non-commercial
license_link: >-
  https://github.com/AnyiRao/segment-anything-with-clip/blob/main/LICENSE
language:
  - en
tags:
  - image-matting
  - transparent-objects
  - alpha-matte
  - trimap
  - computer-vision
  - matting
size_categories:
  - n<1K
task_categories:
  - image-segmentation
pretty_name: Transparent-460
dataset_info:
  features:
    - name: fg
      dtype: image
    - name: alpha
      dtype: image
  splits:
    - name: train
      num_examples: 410
    - name: test
      num_examples: 50
---

# Transparent-460

Transparent object matting dataset introduced in **TransMatting: Enhancing Transparent Objects Matting with Transformers** (ECCV 2022).

Contains 460 transparent foreground images with corresponding alpha mattes. Foregrounds are composited onto background images to generate training/test pairs.

## Dataset Preview

| Foreground (fg) | Alpha Matte | Trimap |
|:-:|:-:|:-:|
| ![fg](https://huggingface.co/datasets/Thinnaphat/transparent-460/resolve/main/Test/fg/al-soot-1Q5LHnMalOM-unsplash.jpg) | ![alpha](https://huggingface.co/datasets/Thinnaphat/transparent-460/resolve/main/Test/alpha/al-soot-1Q5LHnMalOM-unsplash.png) | ![trimap](https://huggingface.co/datasets/Thinnaphat/transparent-460/resolve/main/Test/trimap/al-soot-1Q5LHnMalOM-unsplash.png) |

## Dataset Structure

```
Transparent-460/
├── Train/
│   ├── fg/                              # 410 foreground images (transparent objects)
│   ├── alpha/                           # 410 alpha mattes
│   ├── Composition_code.py             # compositing script
│   ├── transparent-460-train-fg-names.txt
│   └── transparent-460-train-bg-names.txt   # 41,000 COCO BG filenames (100 per FG)
└── Test/
    ├── fg/                              # 50 foreground images
    ├── alpha/                           # 50 alpha mattes
    ├── trimap/                          # 50 trimap masks
    ├── Composition_code.py             # compositing script
    ├── metric_evaluation.py
    ├── transparent-460-test-fg-names.txt
    └── transparent-460-test-bg-names.txt    # 1,000 Pascal VOC BG filenames (20 per FG)
```

### Splits

| Split | FG images | BG source | Composited pairs |
|-------|-----------|-----------|-----------------|
| Train | 410       | COCO train2014 | 41,000 |
| Test  | 50        | Pascal VOC 2007 | 1,000 |

### Data Fields

- **fg** — foreground image of a transparent object (JPEG)
- **alpha** — corresponding alpha matte (PNG, grayscale 0–255)
- **trimap** — ternary region map for test set (PNG; 0=background, 128=unknown, 255=foreground)

## Compositing

Background images are **not** included (COCO / Pascal VOC must be downloaded separately). Use the provided `Composition_code.py` to composite FG images onto BG images:

```python
# Each train FG is composited onto 100 COCO BG images.
# Each test FG is composited onto 20 Pascal VOC BG images.
python Composition_code.py
```

## Example Usage

```python
from datasets import load_dataset

ds = load_dataset("Thinnaphat/transparent-460")

# Train split
train_sample = ds["train"][0]
fg_image  = train_sample["fg"]    # PIL Image
alpha_map = train_sample["alpha"] # PIL Image (grayscale)

# Test split
test_sample = ds["test"][0]
fg_image   = test_sample["fg"]
alpha_map  = test_sample["alpha"]
trimap_map = test_sample["trimap"]
```

## License

**Non-commercial research use only.**

1. Available for non-commercial research purposes only.
2. Images obtained from the Internet; authors not responsible for content.
3. No reproduction, duplicate, copy, sell, trade, or resell for commercial purposes.
4. No further public distribution except internal use within a single organization.
5. Authors reserve the right to terminate access at any time.

## Citation

```bibtex
@inproceedings{cai2022TransMatting,
  title={TransMatting: Enhancing Transparent Objects Matting with Transformers},
  author={Cai, Huanqia and Xue, Fanglei, and Xu, Lele and Guo, Lili},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2022},
}
```