File size: 9,080 Bytes
628d4ae
 
4fc5611
 
 
628d4ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4fc5611
 
 
 
628d4ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4821755
 
 
 
 
 
628d4ae
4fc5611
628d4ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4fc5611
 
 
628d4ae
 
 
 
 
e20c016
 
 
 
 
628d4ae
 
 
 
e20c016
 
 
 
4fc5611
628d4ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4fc5611
 
 
628d4ae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4fc5611
 
 
 
 
628d4ae
 
 
 
 
 
 
 
 
 
 
 
 
4fc5611
628d4ae
 
 
4fc5611
 
 
628d4ae
 
 
321fab0
 
628d4ae
 
321fab0
 
 
 
 
 
 
 
 
628d4ae
 
 
321fab0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
628d4ae
 
 
 
 
 
 
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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
---
pretty_name: Sci-ImageMiner
license: other
license_name: mixed-rights
license_link: LICENSE
task_categories:
- image-classification
- image-to-text
- visual-question-answering
tags:
- scientific-images
- materials-science
- scientific-figures
- atomic-layer-deposition
- atomic-layer-etching
- multimodal
- figure-understanding
size_categories:
- 1K<n<10K
---

# Sci-ImageMiner

Sci-ImageMiner is a scientific-image dataset for multimodal figure
understanding. The initial release focuses on figures extracted from atomic
layer deposition (ALD) and atomic layer etching (ALE) publications in materials
science. The schema is designed to support future scientific domains,
subdomains, image types, data sources, and annotation tasks.

This dataset contains mixed-rights content. Metadata and annotations are
described as CC BY 4.0 in the upstream repository, while source images are
publisher/author copyrighted and are provided for non-commercial research use
only unless the original source license grants broader rights.

## Dataset Purpose

The dataset supports research on scientific image understanding tasks:

- panel-level figure type classification;
- panel localization with bounding boxes;
- panel-level summarization;
- data extraction from plots and scientific visuals;
- visual question answering grounded in scientific figures.

## Initial Scope

The current broad domain is:

```text
materials_science
```

The initial subdomains represented by records are:

```text
atomic_layer_deposition
atomic_layer_etching
```

The current source data also distinguishes:

```text
experimental
simulation
```

These values describe the current release only. Future domains such as
chemistry, biology, physics, earth science, medicine, or engineering are not
included unless records are actually added and curated.

## Splits

The dataset exposes three Hugging Face splits:

| Split | Records |
| --- | ---: |
| `train` | 1,170 |
| `validation` | 201 |
| `test` | 580 |
| **Total** | **1,951** |

The public rows include one `split` field with the Hugging Face split value.

## Dataset Structure

The dataset is packaged in an ImageFolder-compatible layout:

```text
train/
  metadata.jsonl
  images/
validation/
  metadata.jsonl
  images/
test/
  metadata.jsonl
  images/
```

Each metadata row references its image with `file_name`. When loaded with
Hugging Face Datasets, this becomes an `image` column that renders in the
Dataset Viewer.

## Field Definitions

Core fields:

| Field | Description |
| --- | --- |
| `image` | Hugging Face image feature created from `file_name`. |
| `caption` | Source figure caption when available. |
| `id` | Stable dataset-level record id. |
| `sample_id` | Original source annotation `sample_id`. |
| `split` | Hugging Face split. |
| `domain` | Broad scientific domain. |
| `subdomain` | Controlled scientific subdomain. |
| `topic` | Reserved for curated lower-level topics or processes. |
| `study_type` | Experimental or simulation study type. |
| `classification` | Source classification annotations as a list of `{panel_id, label}` entries. |
| `summarization` | Source summarization annotations as a list of `{panel_id, summary}` entries. |
| `data_extraction` | Source data extraction annotations as a list of `{panel_id, data}` entries. |
| `vqa` | Source visual question-answering annotations as a list of `{panel_id, question_type, question, answer_type, answer}` entries. |
| `bbox` | Source bounding boxes as a list of `{panel_id, x, y, width, height}` entries. |
| `source` | Publication and source-file provenance. |
| `provenance` | Extraction and annotation provenance. |
| `schema_version` | Dataset schema version. |

The source annotation fields use panel ids such as `a`, `b`, and `c`. They are
stored as lists rather than arbitrary-key objects so that Hugging Face can infer
a stable viewer schema. If an annotation type does not apply or was not provided
for a figure, it is represented as an empty list.

## Images and Provenance

Images were extracted from published scientific articles. Source PDFs and
`content.json` extraction outputs are present in the upstream repository. The
conversion preserves relative paths to:

- original image files;
- original annotation JSON files;
- source `content.json` files;
- source PDF filenames where available.

Publication metadata such as title, author line, DOI, and publication year is
best-effort and should be reviewed before being treated as authoritative.

## Loading

After upload, users should be able to load the dataset with:

```python
from datasets import load_dataset

dataset = load_dataset("SciKnowOrg/Sci-ImageMiner")
```

For local generated artifacts:

```python
from datasets import load_dataset

dataset = load_dataset("imagefolder", data_dir="hf/dataset")
```

Display an image:

```python
example = dataset["train"][0]
image = example["image"]
image.show()
print(example["caption"])
print(example["classification"])
print(example["bbox"])
print(example["vqa"])
```

## Known Limitations

- The initial release focuses only on ALD and ALE materials-science figures.
- Some records have classification and bounding boxes but no summarization,
  data extraction, or VQA annotations.
- Some captions or DOI fields are missing or best-effort extracted.
- Byte-identical image duplicates exist, including some across original splits.
  Existing split assignments are preserved, but users should consider this when
  using the data as a benchmark.
- The source image redistribution status needs human review before public
  upload.

## Licensing and Redistribution

Annotations and metadata are described in the upstream repository as CC BY 4.0.

Images were extracted from published scientific articles. Copyright remains
with original authors and/or publishers. The dataset repository uses a
mixed-rights license marker rather than a plain CC BY marker because the image
files are not released under CC BY 4.0 by this dataset. Treat source images as
non-commercial research-use-only unless the original publication license grants
broader rights.

Maintainers should decide whether the public dataset must be gated or whether a
metadata-only variant is needed for images whose redistribution rights are not
clear.

## Contributions

Community contributions should follow `hf/CONTRIBUTING_DATASET.md` in the upstream
repository. Contributions must include:

- image files;
- `metadata.jsonl`;
- source publication provenance;
- source-style annotation fields;
- image reuse rights or license information;
- validation output.

New vocabulary values for domains, subdomains, or study types should be proposed
through pull requests to the versioned vocabulary files. Contributors must not
upload images they lack permission to redistribute.

## Citation

The Sci-ImageMiner project vision is described in the following working paper,
pre-released on Zenodo. Please cite this paper if you find the project useful:

```bibtex
@misc{d_souza_2025_17130928,
  author       = {D'Souza, Jennifer},
  title        = {A Pathway to General-Purpose Scientific AI:
                  Multimodal Comprehension of Scientific Images},
  month        = sep,
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.17130928},
  url          = {https://doi.org/10.5281/zenodo.17130928},
}
```

Sci-ImageMiner was also featured as the ICDAR 2026 Competition on Information
Extraction from Atomic Layer Deposition/Etching (ALD/E) Scientific Figures,
organized as part of the ICDAR 2026 competitions. Please cite the competition
report when referring to the benchmark, dataset, competition, or associated
information-extraction tasks:

```bibtex
@article{ahmed2026icdar,
  title   = {ICDAR 2026 Competition on Information Extraction from
             Atomic Layer Deposition/Etching (ALD/E) Scientific Figures},
  author  = {Ahmed, Fahad and Auer, S{\"o}ren and D'Souza, Jennifer},
  journal = {arXiv preprint arXiv:2607.26848},
  year    = {2026},
  url     = {https://arxiv.org/abs/2607.26848}
}
```

## Acknowledgements

Within Sci-ImageMiner, the development of the expert-annotated
ALD/E-ImageMiner benchmark resource was funded by the NFDI4DataScience
initiative, funded by the German Research Foundation (DFG, Grant ID:
460234259) under the Speedboat Annotation Project funding scheme.

This research endeavor is conducted in the context of the AI-Aware Pathways to
Sustainable Semiconductor Process and Manufacturing Technologies (AWASES)
initiative (Mackus et al., 2024), funded by Merck and Intel, with collaboration
between Eindhoven University, Leibniz University Hannover's L3S Research
Centre, and University of Warwick. AWASES hosts three fully funded PhD positions
and supports advances in generative AI, multimodal models, and FAIR scientific
knowledge graph construction.

## Versioning

This draft uses schema version `1.0.0`.

Schema changes should be documented in the dataset card and in release notes.
Additive fields are preferred. Breaking changes should require a new major
schema version or a new dataset configuration.