COREL Category Labels

By design, the images of the COREL collection are organized into 600 classes (100 images per class). Approximately 240 classes represent images of varying category level or themes (e.g. images of Turkey) and are therefore unsuitable to test category performance. But the remaining 360 classes can be assigned to a human category such as superordinate, basic-level and subordinate category. Such a categorization is sometimes ambivalent because many scenes can be assigned or interpreted in various ways as humans possess different entry-level categories, also called perception subjectivity. To avoid a strong perception bias, the classes were categorized by two persons (the author and a research assistant). Preference was given to assignments which follow structural similarity: for instance a toy car was labeled vehicle/bus/toy as superordinate, basic-level and subordinate category respectively, and not toy/vehicle/bus.

The image classes were categorized into 10 superordinate, 112 basic-level and 143 subordinate categories. For that purpose, the COREL image classes were numbered 1 through 600 corresponding to the directory names 0010 through 6010.

[Tex file] containing indices for images requiring 90-deg rotation.

[Tex file] containing indices for images requiring up-side down flipping (in addition to rotation).

[Excel file] containing category labels for the 600 image classes.

The following superordinate labels were chosen for the 360 classes (number of image classes, proportion of the entire collection): animals (68, 11.3%), landscapes (49, 8.2%), persons (48, 8.0%), vehicles (44, 7.3%), parts (30, 5.0%), textures (30, 5.0%), buildings (29, 4.8%), activities (25, 4.2%), plants (22, 3.7%), food (12, 2.0%).

Examples of basic-levels are (in decreasing proportion): wild animals (27, 4.5%), patterns (25, 4.2%), sports (25, 4.2%), flowers (17, 2.8%), aircrafts (16, 2.7%), models (13, 2.2%), birds (11, 1.8%), water animals (10, 1.7%), cars (9, 1.5%), canyons (7, 1.2%), different cultures (7, 1.2%), mountain sceneries (7, 1.2%), ships (7, 1.2%).

Examples of subordinate categories are: fashion models (10, 1.7%), jets (7, 1.2%), seawater fish (7, 1.2%), garden flowers (6, 1.0%), paintings (6, 1.0%), old-timers (5, 0.8%), fractals (4, 0.7%), minerals (4, 0.7%), mountains with snow (4, 0.7%).

The labeling was primarily carried out by Nadine Hartig.