The Object Detection for Olfactory References (ODOR) Dataset

The Object Detection for Olfactory References (ODOR) Dataset

Real-world applications of computer vision in the humanities require algorithms to be robust against artistic abstraction, peripheral objects, and subtle differences between fine-grained target classes. 

Existing datasets provide instance-level annotations on artworks but are generally biased towards the image centre and limited with regard to detailed object classes. The ODOR dataset fills this gap, offering 38,116 object-level annotations across 4,712 images, spanning an extensive set of 139 fine-grained categories. 

It has challenging dataset properties, such as a detailed set of categories, dense and overlapping objects, and spatial distribution over the whole image canvas. 

Inspiring further research on artwork object detection and broader visual cultural heritage studies, the dataset challenges researchers to explore the intersection of object recognition and smell perception.

How to use

The annotations are provided in COCO JSON format. To represent the two-level hierarchy of the object classes, we make use of the supercategory field in the categories array as defined by COCO. In addition to the object-level annotations, we provide an additional CSV file with image-level metadata, which includes content-related fields, such as Iconclass codes or image descriptions, as well as formal annotations, such as artist, license, or creation year. 

In addition to a zip containing the dataset images, we provide links to their source collections in the metadata file and a Python script to conveniently download the artwork images (`download_imgs.py`).

The mapping between the `images` array of the `annotations.json` and the `metadata.csv` file can be accomplished via the `file_name` attribute of the elements of the `images` array and the unique `File Name` column of the `metadata.csv` file, respectively.