Caltech-UCSD Birds 200

Abstract

Caltech-UCSD Birds 200 (CUB-200) is a challenging image dataset annotated with 200 bird species. It was created to enable the study of subordinate categorization, which is not possible with other popular datasets that focus on basic level categories (such as PASCAL VOC, Caltech-101, etc). The images were downloaded from the website Flickr and filtered by workers on Amazon Mechanical Turk. Each image is annotated with a bounding box, a rough bird segmentation, and a set of attribute labels. forehead_color black black black breast_pattern solid solid solid breast_color white white white head_pattern plain capped plain back_color white white black wing_color grey/white grey white leg_color orange orange orange size medium large medium bill_shape needle dagger dagger wing_shape pointed tapered long primary_color white white white forehead_color red red red breast_pattern solid solid breast_color white white/red white head_pattern capped capped capped back_color wing_color white/ black white/ black white/ black white/black white/ black white/black leg_color buff black black size small medium medium bill_shape dagger multicolored allpurpose allpurpose wing_shape pointed tapered pointed primary_color black, red white, black white, black Figure 1: Images and annotations from CUB-200. Each example image is shown with a rough outline (segmentation) in green. To the right of each image is a table of attributes (one per row, 11 out of a total of 25 attributes shown), and attribute-values provided by Amazon Mechanical Turk workers looking at the image. The attribute-values in the three right-most columns in the tables are provided by different workers (across both columns and rows). The font of the attribute-value indicates the confidence of the worker: bold font means the worker was ‘definitely ’ sure of the label, thin means ‘probably’, and grey means ‘guessing’. 1