jaddai commited on
Commit
b5da6ba
·
verified ·
1 Parent(s): 2e0241f

Initial release: curated subset of jaddai/openbrush-75k

Browse files

Filtered slice of jaddai/openbrush-75k (v1.1). Same MIT license, same caption schema, same VLM (Qwen3-VL-30B-A3B). See README.md for details.

README.md ADDED
@@ -0,0 +1,138 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ task_categories:
4
+ - image-to-text
5
+ - text-to-image
6
+ - image-classification
7
+ language:
8
+ - en
9
+ tags:
10
+ - art
11
+ - fine-art
12
+ - paintings
13
+ - public-domain
14
+ - ukiyo-e
15
+ - japanese
16
+ - woodblock
17
+ - movement
18
+ - structured-captions
19
+ - image-captioning
20
+ - diffusion
21
+ - vlm
22
+ size_categories:
23
+ - 1K<n<10K
24
+ configs:
25
+ - config_name: default
26
+ data_files:
27
+ - split: train
28
+ path: data/train-*.parquet
29
+ ---
30
+
31
+ # OpenBrush Ukiyo-e
32
+
33
+ Japanese Ukiyo-e woodblock prints from OpenBrush-75K — the only non-Western style in the parent dataset, separated here for trainers who want a dedicated Japanese woodblock corpus without downloading 75K Western paintings to filter.
34
+
35
+ > **Curated subset of [jaddai/openbrush-75k](https://huggingface.co/datasets/jaddai/openbrush-75k).** Same MIT license, same caption schema, same VLM (Qwen3-VL-30B-A3B). This subset exists so you don't have to download 75,313 images to get to the 1,167 you actually want.
36
+
37
+ ## Why this subset
38
+
39
+ Edo and Meiji-era Japanese woodblock prints. Distinctive flat color, outlined forms, asymmetric composition, and traditional Japanese subject matter (landscapes, kabuki actors, beauties, scenes from daily life).
40
+
41
+ **Filter equivalent on parent:** `style == "Ukiyo-e"`
42
+
43
+ ## Statistics
44
+
45
+ - **Total images:** 1,167
46
+ - **Source:** OpenBrush-75K v1.1 (filtered)
47
+ - **License:** MIT
48
+ - **Caption schema:** v2 structured (TAGS / SUBJECT / ACTION / SETTING / MOOD / STYLE / LIGHTING / COLOR / COMPOSITION)
49
+
50
+
51
+ ## Caption schema
52
+
53
+ Same as the parent. Each row has `caption_full` plus 9 individual fields:
54
+
55
+ ```
56
+ **TAGS:** comma-separated standardized descriptive tags
57
+ **SUBJECT:** Description of the main subject
58
+ **ACTION:** Movement, gesture, what is happening
59
+ **SETTING:** Environment and context
60
+ **MOOD:** Emotional tone and atmosphere
61
+ **STYLE:** Artistic technique and stylistic analysis
62
+ **LIGHTING:** Direction, quality, temperature
63
+ **COLOR:** Palette and color relationships
64
+ **COMPOSITION:** Framing, balance, focal points
65
+ ```
66
+
67
+ ## Schema
68
+
69
+ | Field | Type | Description |
70
+ |-------|------|-------------|
71
+ | `id` | string | Unique identifier |
72
+ | `image` | Image | The artwork (PNG bytes) |
73
+ | `width` | int | Image width in pixels |
74
+ | `height` | int | Image height in pixels |
75
+ | `artist` | string | Artist name |
76
+ | `style` | string | Art movement/style |
77
+ | `genre` | string | Genre |
78
+ | `tags` | list[string] | Standardized descriptive tags |
79
+ | `subject`, `action`, `setting`, `mood`, `style_description`, `lighting`, `color`, `composition` | string | Per-aspect caption fields |
80
+ | `caption_full` | string | Full structured caption |
81
+ | `source_file` | string | Original filename |
82
+
83
+ ## Loading
84
+
85
+ ```python
86
+ from datasets import load_dataset
87
+
88
+ dataset = load_dataset("jaddai/openbrush-ukiyo-e")
89
+ print(len(dataset["train"])) # 1,167
90
+
91
+ example = dataset["train"][0]
92
+ print(example["caption_full"])
93
+ ```
94
+
95
+ ## Relationship to the parent dataset
96
+
97
+ This is a filtered slice of [jaddai/openbrush-75k](https://huggingface.co/datasets/jaddai/openbrush-75k) v1.1. All images, captions, and metadata are identical to those in the parent — no recaptioning or modification.
98
+
99
+ If you want the **complete** OpenBrush corpus (75,313 paintings), use the parent. If you want **just this slice**, use this subset to save bandwidth and disk.
100
+
101
+ ### Sibling subsets
102
+
103
+ - [openbrush-van-gogh](https://huggingface.co/datasets/jaddai/openbrush-van-gogh)
104
+ - [openbrush-monet](https://huggingface.co/datasets/jaddai/openbrush-monet)
105
+ - [openbrush-renoir](https://huggingface.co/datasets/jaddai/openbrush-renoir)
106
+ - [openbrush-rembrandt](https://huggingface.co/datasets/jaddai/openbrush-rembrandt)
107
+ - [openbrush-impressionism](https://huggingface.co/datasets/jaddai/openbrush-impressionism)
108
+ - [openbrush-renaissance](https://huggingface.co/datasets/jaddai/openbrush-renaissance)
109
+ - [openbrush-ukiyo-e](https://huggingface.co/datasets/jaddai/openbrush-ukiyo-e)
110
+ - [openbrush-baroque](https://huggingface.co/datasets/jaddai/openbrush-baroque)
111
+ - [openbrush-landscapes](https://huggingface.co/datasets/jaddai/openbrush-landscapes)
112
+ - [openbrush-portraits](https://huggingface.co/datasets/jaddai/openbrush-portraits)
113
+ - [openbrush-religious-art](https://huggingface.co/datasets/jaddai/openbrush-religious-art)
114
+ - [openbrush-anonymous-masters](https://huggingface.co/datasets/jaddai/openbrush-anonymous-masters)
115
+ - [openbrush-impressionist-landscapes](https://huggingface.co/datasets/jaddai/openbrush-impressionist-landscapes)
116
+
117
+ ## License
118
+
119
+ MIT. Underlying artworks are public domain (all artists died before 1954). Structured captions are original works released under MIT.
120
+
121
+ ## Citation
122
+
123
+ ```bibtex
124
+ @dataset{openbrush_openbrush_ukiyo_e,
125
+ title={OpenBrush Ukiyo-e},
126
+ author={jaddai},
127
+ year={2026},
128
+ publisher={Hugging Face},
129
+ url={https://huggingface.co/datasets/jaddai/openbrush-ukiyo-e},
130
+ note={Curated subset of openbrush-75k}
131
+ }
132
+ ```
133
+
134
+ ## Acknowledgments
135
+
136
+ - Original artwork images derived from public domain art collections
137
+ - Captions generated using Qwen3-VL-30B-A3B
138
+ - Tag standardization assisted by LLM-based semantic mapping
data/train-00000-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48d438e4efae3059b2cffb0e8ab13698c4bb64d0c0f11a8abef053f8c52f6375
3
+ size 424884953
data/train-00001-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ade94e33683ab26ae32e79d4744e016de8398dbd9099b78075f933b1e36c680c
3
+ size 390875259
data/train-00002-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d7cd87f9d54d50405677670a46e1880019bb6cad0ca93219b874f44cbc7d522b
3
+ size 392240531
data/train-00003-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:88a8c0643654bd7c747c6899c9a8f77a146566163ba1ae3e546343864e12e7ff
3
+ size 415565754
data/train-00004-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b5d7b28cdb5228c0c2633ecb52b613d08a79940912a2e89176bc3a1564744f4
3
+ size 402919083
data/train-00005-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ff8a0f6bb5b7c9c02a57dec1148c0d4e022cf9f88561bb3dd39901980ee715ce
3
+ size 412664139
data/train-00006-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d588773d47141a2fbe53983fdd649d387166996646081dbf89bedba1e01042e
3
+ size 389261898
data/train-00007-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:415d072674b7115998c57ac6987b77bb925252b8c92ddadb8e43c7ae1188d156
3
+ size 435732663
data/train-00008-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5356a8c7455a000f9879b72426aa3e55db1ce1d5b682c8ac8f7bfeda6d21bf4b
3
+ size 418149112
data/train-00009-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:00ceb6af00ef81a1e6fb228ad3571350a701282163199705bce798d828cf96cb
3
+ size 471287302
data/train-00010-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bab27851861759cc27928f6378454e6c74b645c8f3abc7c14f1d4ed71bbe7b45
3
+ size 457897784
data/train-00011-of-00012.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1cde1ff0b66f882352526f45304f5500b37a1f85dcddc65133a6827ee68a35a7
3
+ size 337225175
dataset_info.json ADDED
@@ -0,0 +1,14 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "dataset_name": "openbrush-ukiyo-e",
3
+ "version": "1.0.0",
4
+ "license": "MIT",
5
+ "parent_dataset": "jaddai/openbrush-75k",
6
+ "splits": {
7
+ "train": {
8
+ "num_examples": 1167,
9
+ "num_shards": 12
10
+ }
11
+ },
12
+ "title": "OpenBrush Ukiyo-e",
13
+ "summary": "Japanese Ukiyo-e woodblock prints from OpenBrush-75K \u2014 the only non-Western style in the parent dataset, separated here for trainers who want a dedicated Japanese woodblock corpus without downloading 75K Western paintings to filter."
14
+ }