trojblue commited on
Commit
873d996
·
verified ·
1 Parent(s): 8445b5c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +53 -169
README.md CHANGED
@@ -141,200 +141,79 @@ pretty_name: Danbooru 2025 Metadata
141
  size_categories:
142
  - 1M<n<10M
143
  ---
144
-
145
  # Dataset Card for Danbooru 2025 Metadata
146
 
147
- **Current max id: `8,877,698` (Feb.18, 2025)**
148
 
149
- This dataset repo provides comprehensive, up-to-date metadata for the Danbooru booru site. All metadata was freshly scraped starting on January 2, 2025, resulting in more extensive tag annotations for older posts, fewer errors, and reduced occurrences of non-labelled AI-generated images in the data.
150
 
151
  ## Dataset Details
152
 
153
- **What is this?**
154
- A refreshed, Parquet-formatted metadata dump of Danbooru, current as of Feb.18, 2025.
155
-
156
-
157
- **Why choose this over other Danbooru scrapes?**
158
 
159
- The dataset includes all available metadata in one place, eliminating the need to gather and merge data from multiple sources manually.
160
 
161
- It features more annotations and fewer untagged or mislabeled AI-generated images than older scrapes. Additionally, historical tag renames and additions are accurately reflected, ensuring easier and more reliable downstream use.
 
 
162
 
 
163
 
164
- ## Uses
165
-
166
- The dataset can be loaded or filtered with the Huggingface `datasets` library:
167
 
168
  ```python
169
- from datasets import Dataset, load_dataset
170
 
171
- danbooru_dataset = load_dataset("trojblue/danbooru2025-metadata", split="train")
172
- df = danbooru_dataset.to_pandas()
173
  ```
174
 
 
175
 
176
  ## Dataset Structure
177
 
178
- For better compatability, the columns are converted from danbooru jsons with minimal change:
179
 
180
  ```
181
- Index(['approver_id', 'bit_flags', 'created_at', 'down_score', 'fav_count',
182
- 'file_ext', 'file_size', 'file_url', 'has_active_children',
183
- 'has_children', 'has_large', 'has_visible_children', 'id',
184
- 'image_height', 'image_width', 'is_banned', 'is_deleted', 'is_flagged',
185
- 'is_pending', 'large_file_url', 'last_comment_bumped_at',
186
- 'last_commented_at', 'last_noted_at', 'md5', 'media_asset_created_at',
187
- 'media_asset_duration', 'media_asset_file_ext', 'media_asset_file_key',
188
- 'media_asset_file_size', 'media_asset_id', 'media_asset_image_height',
189
- 'media_asset_image_width', 'media_asset_is_public', 'media_asset_md5',
190
- 'media_asset_pixel_hash', 'media_asset_status',
191
- 'media_asset_updated_at', 'media_asset_variants', 'parent_id',
192
- 'pixiv_id', 'preview_file_url', 'rating', 'score', 'source',
193
- 'tag_count', 'tag_count_artist', 'tag_count_character',
194
- 'tag_count_copyright', 'tag_count_general', 'tag_count_meta',
195
- 'tag_string', 'tag_string_artist', 'tag_string_character',
196
- 'tag_string_copyright', 'tag_string_general', 'tag_string_meta',
197
- 'up_score', 'updated_at', 'uploader_id'],
198
- dtype='object')
199
  ```
200
 
201
- <div>
202
- <style scoped>
203
- .dataframe tbody tr th:only-of-type {
204
- vertical-align: middle;
205
- }
206
-
207
- .dataframe tbody tr th {
208
- vertical-align: top;
209
- }
210
-
211
- .dataframe thead th {
212
- text-align: right;
213
- }
214
- </style>
215
- <table border="1" class="dataframe">
216
- <thead>
217
- <tr style="text-align: right;">
218
- <th></th>
219
- <th>approver_id</th>
220
- <th>bit_flags</th>
221
- <th>created_at</th>
222
- <th>down_score</th>
223
- <th>fav_count</th>
224
- <th>file_ext</th>
225
- <th>file_size</th>
226
- <th>file_url</th>
227
- <th>has_active_children</th>
228
- <th>has_children</th>
229
- <th>...</th>
230
- <th>tag_count_meta</th>
231
- <th>tag_string</th>
232
- <th>tag_string_artist</th>
233
- <th>tag_string_character</th>
234
- <th>tag_string_copyright</th>
235
- <th>tag_string_general</th>
236
- <th>tag_string_meta</th>
237
- <th>up_score</th>
238
- <th>updated_at</th>
239
- <th>uploader_id</th>
240
- </tr>
241
- </thead>
242
- <tbody>
243
- <tr>
244
- <th>0</th>
245
- <td>NaN</td>
246
- <td>0</td>
247
- <td>2015-08-07T23:23:45.072-04:00</td>
248
- <td>0</td>
249
- <td>66</td>
250
- <td>jpg</td>
251
- <td>4134797</td>
252
- <td>https://cdn.donmai.us/original/a1/b3/a1b3d0fa9...</td>
253
- <td>False</td>
254
- <td>False</td>
255
- <td>...</td>
256
- <td>3</td>
257
- <td>1girl absurdres ass bangle bikini black_bikini...</td>
258
- <td>kyouka.</td>
259
- <td>marie_(splatoon)</td>
260
- <td>splatoon_(series) splatoon_1</td>
261
- <td>1girl ass bangle bikini black_bikini blush bra...</td>
262
- <td>absurdres commentary_request highres</td>
263
- <td>15</td>
264
- <td>2024-06-25T15:32:44.291-04:00</td>
265
- <td>420773</td>
266
- </tr>
267
- <tr>
268
- <th>1</th>
269
- <td>NaN</td>
270
- <td>0</td>
271
- <td>2008-03-05T01:52:28.194-05:00</td>
272
- <td>0</td>
273
- <td>7</td>
274
- <td>jpg</td>
275
- <td>380323</td>
276
- <td>https://cdn.donmai.us/original/d6/10/d6107a13b...</td>
277
- <td>False</td>
278
- <td>False</td>
279
- <td>...</td>
280
- <td>2</td>
281
- <td>1girl aqua_hair bad_id bad_pixiv_id guitar hat...</td>
282
- <td>shimeko</td>
283
- <td>hatsune_miku</td>
284
- <td>vocaloid</td>
285
- <td>1girl aqua_hair guitar instrument long_hair so...</td>
286
- <td>bad_id bad_pixiv_id</td>
287
- <td>4</td>
288
- <td>2018-01-23T00:32:10.080-05:00</td>
289
- <td>1309</td>
290
- </tr>
291
- <tr>
292
- <th>2</th>
293
- <td>85307.0</td>
294
- <td>0</td>
295
- <td>2015-08-07T23:26:12.355-04:00</td>
296
- <td>0</td>
297
- <td>10</td>
298
- <td>jpg</td>
299
- <td>208409</td>
300
- <td>https://cdn.donmai.us/original/a1/2c/a12ce629f...</td>
301
- <td>False</td>
302
- <td>False</td>
303
- <td>...</td>
304
- <td>1</td>
305
- <td>1boy 1girl blush boots carrying closed_eyes co...</td>
306
- <td>yuuryuu_nagare</td>
307
- <td>jon_(pixiv_fantasia_iii) race_(pixiv_fantasia)</td>
308
- <td>pixiv_fantasia pixiv_fantasia_3</td>
309
- <td>1boy 1girl blush boots carrying closed_eyes da...</td>
310
- <td>commentary_request</td>
311
- <td>3</td>
312
- <td>2022-05-25T02:26:06.588-04:00</td>
313
- <td>95963</td>
314
- </tr>
315
- </tbody>
316
- </table>
317
- </div>
318
-
319
-
320
  ## Dataset Creation
321
 
322
- We scraped all post IDs on Danbooru from 1 up to the latest. Some restricted tags (e.g. `loli`) were hidden by the site and require a gold account to access, so they are not present.
323
- For a more complete (but older) metadata reference, you may wish to combine this with Danbooru2021 or similar previous scrapes.
 
 
 
324
 
325
- The scraping process used a pool of roughly 400 IPs over six hours, ensuring consistent tag definitions. Below is a simplified example of the process used to convert the metadata into Parquet:
326
 
327
  ```python
328
  import pandas as pd
329
  from pandarallel import pandarallel
330
 
331
- # Initialize pandarallel
332
  pandarallel.initialize(nb_workers=4, progress_bar=True)
333
 
334
  def flatten_dict(d, parent_key='', sep='_'):
335
- """
336
- Flattens a nested dictionary.
337
- """
338
  items = []
339
  for k, v in d.items():
340
  new_key = f"{parent_key}{sep}{k}" if parent_key else k
@@ -347,22 +226,27 @@ def flatten_dict(d, parent_key='', sep='_'):
347
  return dict(items)
348
 
349
  def extract_all_illust_info(json_content):
350
- """
351
- Parses and flattens Danbooru JSON into a pandas Series.
352
- """
353
  flattened_data = flatten_dict(json_content)
354
  return pd.Series(flattened_data)
355
 
356
  def dicts_to_dataframe_parallel(dicts):
357
- """
358
- Converts a list of dicts to a flattened DataFrame using pandarallel.
359
- """
360
  df = pd.DataFrame(dicts)
361
  flattened_df = df.parallel_apply(lambda row: extract_all_illust_info(row.to_dict()), axis=1)
362
  return flattened_df
363
  ```
364
 
 
 
 
 
 
 
 
 
 
365
 
366
- ### Recommendations
367
 
368
- Users should be aware of potential biases and limitations, including the presence of adult content in some tags. More details and mitigations may be needed.
 
141
  size_categories:
142
  - 1M<n<10M
143
  ---
 
144
  # Dataset Card for Danbooru 2025 Metadata
145
 
146
+ **Latest Post ID**: 8,877,698 (as of February 18, 2025)
147
 
148
+ This repository provides a comprehensive, up-to-date metadata dump for Danbooru. The metadata was freshly scraped starting January 2, 2025, featuring more extensive tag annotations for older posts, fewer errors, and fewer unlabeled AI-generated images compared to previous scrapes.
149
 
150
  ## Dataset Details
151
 
152
+ **Overview**
153
+ Danbooru is a well-known imageboard focusing on anime-style artwork, hosting millions of user-submitted images with extensive tagging. This dataset offers metadata (in Parquet format) for all posts up to the specified date, including details such as tags, upload timestamps, and file properties.
 
 
 
154
 
155
+ **Key Advantages**
156
 
157
+ - **Consolidated Metadata**: All available metadata is contained within this single dataset, eliminating the need to merge multiple partial scrapes.
158
+ - **Improved Tag Accuracy**: Historical tag renames and additions are accurately reflected, reducing the potential mismatch or redundancy often found in older metadata dumps.
159
+ - **Less AI Noise**: Compared to many legacy scrapes, the 2025 data incorporates updated annotations and filters out many unlabeled AI-generated images.
160
 
161
+ ## Usage
162
 
163
+ You can load and filter this dataset using the Hugging Face `datasets` library:
 
 
164
 
165
  ```python
166
+ from datasets import load_dataset
167
 
168
+ danbooru_metadata = load_dataset("trojblue/danbooru2025-metadata", split="train")
169
+ df = danbooru_metadata.to_pandas()
170
  ```
171
 
172
+ This metadata can be used for research, indexing, or as a foundation for building image-based machine learning pipelines. However, please be mindful of any copyright, content, or platform-specific policies.
173
 
174
  ## Dataset Structure
175
 
176
+ The metadata schema is closely aligned with Danbooru’s JSON structure, ensuring familiarity for those who have used other Danbooru scrapes. Below are the main columns:
177
 
178
  ```
179
+ Index([
180
+ 'approver_id', 'bit_flags', 'created_at', 'down_score', 'fav_count',
181
+ 'file_ext', 'file_size', 'file_url', 'has_active_children', 'has_children',
182
+ 'has_large', 'has_visible_children', 'id', 'image_height', 'image_width',
183
+ 'is_banned', 'is_deleted', 'is_flagged', 'is_pending', 'large_file_url',
184
+ 'last_comment_bumped_at', 'last_commented_at', 'last_noted_at', 'md5',
185
+ 'media_asset_created_at', 'media_asset_duration', 'media_asset_file_ext',
186
+ 'media_asset_file_key', 'media_asset_file_size', 'media_asset_id',
187
+ 'media_asset_image_height', 'media_asset_image_width',
188
+ 'media_asset_is_public', 'media_asset_md5', 'media_asset_pixel_hash',
189
+ 'media_asset_status', 'media_asset_updated_at', 'media_asset_variants',
190
+ 'parent_id', 'pixiv_id', 'preview_file_url', 'rating', 'score', 'source',
191
+ 'tag_count', 'tag_count_artist', 'tag_count_character',
192
+ 'tag_count_copyright', 'tag_count_general', 'tag_count_meta', 'tag_string',
193
+ 'tag_string_artist', 'tag_string_character', 'tag_string_copyright',
194
+ 'tag_string_general', 'tag_string_meta', 'up_score', 'updated_at',
195
+ 'uploader_id'
196
+ ], dtype='object')
197
  ```
198
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
199
  ## Dataset Creation
200
 
201
+ **Scraping Process**
202
+
203
+ - Post IDs from 1 to the latest ID (8,877,698) were retrieved using a distributed scraping approach.
204
+ - Certain restricted tags (e.g., `loli`) are inaccessible without special permissions and are therefore absent in this dataset.
205
+ - If you require more comprehensive metadata (including hidden or restricted tags), consider merging this data with older scrapes such as Danbooru2021.
206
 
207
+ Below is a simplified example of how the raw JSON was converted into a flattened Parquet file:
208
 
209
  ```python
210
  import pandas as pd
211
  from pandarallel import pandarallel
212
 
 
213
  pandarallel.initialize(nb_workers=4, progress_bar=True)
214
 
215
  def flatten_dict(d, parent_key='', sep='_'):
216
+ """Recursively flattens a nested dictionary."""
 
 
217
  items = []
218
  for k, v in d.items():
219
  new_key = f"{parent_key}{sep}{k}" if parent_key else k
 
226
  return dict(items)
227
 
228
  def extract_all_illust_info(json_content):
229
+ """Parses and flattens Danbooru JSON into a pandas Series."""
 
 
230
  flattened_data = flatten_dict(json_content)
231
  return pd.Series(flattened_data)
232
 
233
  def dicts_to_dataframe_parallel(dicts):
234
+ """Converts a list of dicts to a flattened DataFrame using pandarallel."""
 
 
235
  df = pd.DataFrame(dicts)
236
  flattened_df = df.parallel_apply(lambda row: extract_all_illust_info(row.to_dict()), axis=1)
237
  return flattened_df
238
  ```
239
 
240
+ ## Considerations & Recommendations
241
+
242
+ - **Adult/NSFW Content**: Danbooru includes adult imagery and explicit tags. Exercise caution, especially if sharing or using this data in public-facing contexts.
243
+ - **Licensing & Copyright**: Images referenced by this metadata may be copyrighted. Refer to Danbooru’s Terms of Service and respect artists’ rights.
244
+ - **Potential Bias**: Tags are community-curated and can reflect the inherent biases of the user base (e.g., under- or over-tagging certain categories).
245
+ - **Missing or Restricted Tags**: Some tags require special permissions on Danbooru; hence they do not appear in this dataset.
246
+
247
+ For further integration of historical data, consider merging with previous Danbooru scrapes.
248
+ If you use this dataset in research or production, please cite appropriately and abide by all relevant terms and conditions.
249
 
250
+ ------
251
 
252
+ *Last Updated: February 18, 2025*