File size: 4,053 Bytes
076a51a
 
 
 
 
 
 
 
 
 
 
 
 
 
fd8bc79
076a51a
 
 
 
14dd349
076a51a
 
 
 
 
 
 
 
 
 
 
 
 
 
14dd349
076a51a
 
d5d8263
076a51a
d5d8263
14dd349
d5d8263
 
 
 
 
 
076a51a
d5d8263
076a51a
d5d8263
076a51a
d5d8263
 
 
 
14dd349
d5d8263
076a51a
d5d8263
 
 
 
 
 
076a51a
d5d8263
076a51a
14dd349
 
d5d8263
076a51a
d5d8263
 
 
076a51a
14dd349
076a51a
64a3d01
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
076a51a
 
14dd349
 
076a51a
 
 
 
 
 
 
 
fd8bc79
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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
---
license: cc-by-nc-4.0
task_categories:
- image-classification
language:
- en
tags:
- lithology
- geology
- rock
- drill core
- core images
- lithology identification
- lithology classification
- DCID
pretty_name: Drill Core Image Dataset (DCID)
size_categories:
- 10K<n<100K
---

# Dataset Card for Drill Core Image Dataset (DCID)

## Dataset Details

### Dataset Description

The Drill Core Image Dataset (DCID) is a large-scale benchmark designed for lithology classification based on RGB core images. It provides two primary versions:

- **DCID-7**: 7 lithology categories with 5,000 images per class.
- **DCID-35**: 35 lithology categories with 1,000 images per class.

All original images are 512×512 pixels in resolution. Each category is split into training and testing subsets in an 8:2 ratio. Additional variants are generated by resizing to smaller resolutions (32, 64, 128, 256) and applying real-world data augmentation (RWDA) to simulate image imperfections.

- **Curated by:** Jia-Yu Li, Ji-Zhou Tang, et al.
- **Shared by:** Jia-Yu Li (lijiayu1120@tongji.edu.cn), Ji-Zhou Tang (jeremytang@tongji.edu.cn)
- **License:** CC BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0)

---

### Visual Overview

#### DCID Naming Convention
The dataset naming follows the **DCID-R-C-L-I** format:
- `R`: resolution (32, 64, 128, 256, 512)  
- `C`: number of categories (7 or 35)  
- `L`: RWDA level (0.0 – 0.4)  
- `I`: injection scope (`N`, `T`, `E`, `A`)  

![DCID Naming Convention](./DCID-R-C-L-I.jpg)

---

#### DCID-7 Dataset
The **DCID-7** dataset contains 35,000 images (5,000 per category).  
Each class has 4,000 training and 1,000 testing images (8:2 ratio).  
This version is suitable for evaluating model upper-bound performance.

![DCID-7 Dataset Overview](./DCID-7.jpg)

---

#### DCID-35 Dataset
The **DCID-35** dataset contains 35,000 images (1,000 per category).  
Each class has 800 training and 200 testing images.  
This fine-grained version is designed to assess model generalization under complex conditions.

![DCID-35 Dataset Overview](./DCID-35.jpg)

---

### Dataset Sources

- **GitHub Repository:** [https://github.com/JiayuLi1120/drill-core-image-dataset](https://github.com/JiayuLi1120/drill-core-image-dataset)  
- **Hugging Face Dataset:** [https://huggingface.co/datasets/168sir/drill-core-image-dataset](https://huggingface.co/datasets/168sir/drill-core-image-dataset)  
- **Paper:** [https://doi.org/10.1016/j.petsci.2025.04.013](https://doi.org/10.1016/j.petsci.2025.04.013)

---

## Usage

### Step 1: Download and extract
Download the `DCID.zip` archive from [Hugging Face](https://huggingface.co/datasets/168sir/drill-core-image-dataset) and extract it:

```bash
unzip DCID.zip -d ./DCID
````

This will give you the following folders:

* `DCID-512-7/` and `noise-512-7/`
* `DCID-512-35/` and `noise-512-35/`

---

### Step 2: Build custom dataset versions

We provide a script **`build_dcid_dataset.py`** to generate different dataset variants.

Example: Create a **32×32 resolution, 7 classes, 40% RWDA (train set only)** dataset:

```bash
python build_dcid_dataset.py \
    --root ./DCID \
    --R 32 \
    --C 7 \
    --L 0.4 \
    --I T \
    --out_dir ./output
```

This generates a new dataset at:

```
./output/DCID-32-7-0.4-T/
```

---

### Script Parameters

* **`R`**: target resolution (32, 64, 128, 256)
* **`C`**: number of categories (7 or 35)
* **`L`**: RWDA level (0.0–0.4)
* **`I`**: injection scope:

  * `N`: none
  * `T`: train set only
  * `E`: test set only
  * `A`: all (train + test)

---

## Citation

If you use this dataset in your work, please cite:

```bibtex
@article{Li2025DCID,
  title = {A large-scale, high-quality dataset for lithology identification: Construction and applications},
  author = {Jia-Yu Li and Ji-Zhou Tang and Xian-Zheng Zhao and Bo Fan and Wen-Ya Jiang and Shun-Yao Song and Jian-Bing Li and Kai-Da Chen and Zheng-Guang Zhao},
  journal = {Petroleum Science},
  year = {2025},
  issn = {1995-8226},
  doi = {10.1016/j.petsci.2025.04.013}
}