draft readme

README.md

@@ -287,20 +287,76 @@ size_categories:
 
 ### Dataset Summary
 
-> Large datasets that were made publicly available to the research community over the last 20 years have been a key enabling factor for the advances in deep learning algorithms for NLP or computer vision. These datasets are generally pairs of aligned image / manually annotated metadata, where images are photographs of everyday life. Scholarly and historical content, on the other hand, treat subjects that are not necessarily popular to a general audience, they may not always contain a large number of data points, and new data may be difficult or impossible to collect. Some exceptions do exist, for instance, scientific or health data, but this is not the case for cultural heritage (CH). The poor performance of the best models in computer vision - when tested over artworks - coupled with the lack of extensively annotated datasets for CH, and the fact that artwork images depict objects and actions not captured by photographs, indicate that a CH-specific dataset would be highly valuable for this community. We propose DEArt, at this point primarily an object detection and pose classification dataset meant to be a reference for paintings between the XIIth and the XVIIIth centuries. It contains more than 15000 images, about 80% non-iconic, aligned with manual annotations for the bounding boxes identifying all instances of 69 classes as well as 12 possible poses for boxes identifying human-like objects. Of these, more than 50 classes are CH-specific and thus do not appear in other datasets; these reflect imaginary beings, symbolic entities and other categories related to art. Additionally, existing datasets do not include pose annotations. Our results show that object detectors for the cultural heritage domain can achieve a level of precision comparable to state-of-art models for generic images via transfer learning.
+> DEArt is an object detection and pose classification dataset meant to be a reference for paintings between the XIIth and the XVIIIth centuries. It contains more than 15000 images, about 80% non-iconic, aligned with manual annotations for the bounding boxes identifying all instances of 69 classes as well as 12 possible poses for boxes identifying human-like objects. Of these, more than 50 classes are cultural heritage specific and thus do not appear in other datasets; these reflect imaginary beings, symbolic entities and other categories related to art.
 
 ### Supported Tasks and Leaderboards
 
-[More Information Needed]
+- `object-detection`: This dataset can be used to train or evaluate models for object-detection on historical document images.
+- `image-classification`: This dataset can be used for image classification tasks by using only the labels and not the bounding box information
 
 
 ## Dataset Structure
 
+This dataset has two configurations. These configurations both cover the same data and annotations but provide these annotations in different forms to make it easier to integrate the data with existing processing pipelines. 
+
+- The first configuration, `raw, uses the data's original format. 
+- The second configuration converts the annotations into a format that is closer to the `COCO` annotation format. This is done to make it easier to work with the [`image_processors`](https://huggingface.co/docs/transformers/main_classes/image_processor) (formerly known as`feature_extractor`s) from the `Transformers` models for object detection, which expects data to be in a COCO-style format. 
+
 ### Data Instances
 
-[More Information Needed]
+An instance from the `raw` config:
+```python
+{'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1019x1680>,
+ 'source': 'Europeana Collection',
+ 'width': 1019,
+ 'height': 1680,
+ 'dept': 3,
+ 'segmented': None,
+ 'objects': [{'name': 40,
+   'pose': 3,
+   'diffult': 0,
+   'xmin': 259.0,
+   'ymin': 166.0,
+   'xmax': 679.0,
+   'ymax': 479.0},
+  {'name': 19,
+   'pose': 2,
+   'diffult': 0,
+   'xmin': 115.0,
+   'ymin': 354.0,
+   'xmax': 882.0,
+   'ymax': 1168.0},
+  {'name': 15,
+   'pose': 3,
+   'diffult': 0,
+   'xmin': 445.0,
+   'ymin': 1170.0,
+   'xmax': 579.0,
+   'ymax': 1302.0},
+  {'name': 51,
+   'pose': 3,
+   'diffult': 0,
+   'xmin': 354.0,
+   'ymin': 1196.0,
+   'xmax': 445.0,
+   'ymax': 1330.0},
+  {'name': 51,
+   'pose': 3,
+   'diffult': 0,
+   'xmin': 580.0,
+   'ymin': 1203.0,
+   'xmax': 701.0,
+   'ymax': 1326.0},
+  {'name': 57,
+   'pose': 3,
+   'diffult': 0,
+   'xmin': 203.0,
+   'ymin': 642.0,
+   'xmax': 882.0,
+   'ymax': 1172.0}]}
+```
 
-COCO 
+An instance from the `coco` config:
 
 ```python
 {'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1019x1680>,
@@ -354,16 +410,50 @@ COCO
 
 ### Data Splits
 
-[More Information Needed]
+The dataset doesn't define set splits, so only a train split is provided. The paper associated with the dataset does discuss a train and validation split, but it doesn't appear this split was shared.
 
 ## Dataset Creation
 
+
 ### Curation Rationale
+The creators of the dataset authors outline some of their motivations for creating the dataset in the abstract for their paper:
+
+>Large datasets that were made publicly available to the research community over the last 20 years have been a key enabling factor for the advances in deep learning algorithms for NLP or computer vision. These datasets are generally pairs of aligned image / manually annotated metadata, where images are photographs of everyday life. Scholarly and historical content, on the other hand, treat subjects that are not necessarily popular to a general audience, they may not always contain a large number of data points, and new data may be difficult or impossible to collect. Some exceptions do exist, for instance, scientific or health data, but this is not the case for cultural heritage (CH). The poor performance of the best models in computer vision - when tested over artworks - coupled with the lack of extensively annotated datasets for CH, and the fact that artwork images depict objects and actions not captured by photographs, indicate that a CH-specific dataset would be highly valuable for this community. We propose DEArt, at this point primarily an object detection and pose classification dataset meant to be a reference for paintings between the XIIth and the XVIIIth centuries. It contains more than 15000 images, about 80% non-iconic, aligned with manual annotations for the bounding boxes identifying all instances of 69 classes as well as 12 possible poses for boxes identifying human-like objects. Of these, more than 50 classes are CH-specific and thus do not appear in other datasets; these reflect imaginary beings, symbolic entities and other categories related to art. Additionally, existing datasets do not include pose annotations. 
 
-[More Information Needed]
 
 ### Source Data
 
+The source data comes from several cultural heritage institutions that have shared openly licenced images. The dictionary below shows the institutions and the frequency with which they are the provider of images in the dataset.
+
+```python
+{'National Museum in Warsaw': 2030,
+ 'Europeana Collection': 1991,
+ 'The Art Institute of Chicago': 1237,
+ 'The Metropolitan Museum of Art': 1218,
+ 'Rijksmuseum': 1066,
+ 'National Gallery of Art': 871,
+ 'Philadelphia Museum of Art': 774,
+ 'WikiArt': 687,
+ 'National museum in Krakow': 661,
+ 'National Gallery of Denmark': 660,
+ 'British Museum': 618,
+ 'Victoria and Albert Museum': 561,
+ 'Paul Mellon Centre': 391,
+ 'National Gallery of Scotland': 384,
+ 'Yale University Art Gallery': 376,
+ 'Museo Nacional Thyssen-Bornemisza': 332,
+ 'Harvard Art Museum': 279,
+ 'The National Museum of Norvay': 270,
+ 'LACMA': 262,
+ 'The Cleveland Museum of Art': 242,
+ 'The Leiden Collection': 159,
+ 'The Clark Museum': 77,
+ 'Pharos': 6,
+ 'Wikimedia Commons': 2,
+ 'Wikipedia': 1,
+ 'Unknown': 1}
+```
+
 #### Initial Data Collection and Normalization
 
 [More Information Needed]