COSMOROE Newspaper Caricatures Dataset

The dataset comprises 50 greek newspaper caricatures of the caricaturist Yiannis Ioannou pertaining to politics and environmental issues with gold annotations along the following dimensions:

(a) scene and graphic text recognition, (b) object (inanimate entity) recognition (trajectory of object contour as list of point coordinates), (c) agent (animate entity) recognition  (trajectory of agent contour as list of point coordinates), (d) visual scene recognition (distinction into backaground vs foreground), (e) implied activity identification (gesture, body-movement and object movement, distinction into  "by own force"or "by external force"), (f) body movement and gesture complement identification     (agent, tool, affected object, location of the action), (g) cross-media semantic relation identification (i.e., semantic relations holding between verbal and visual elements employing the COSMOROE (CMR) framework, see related documentation in Pastra 2008, Pastra 2015); (h) all visual elements are tagged/labeled (verbal categorization in Greek). 

Furthermore, the dataset includes an ebook with the following information per caricature: (i) lists of visual and verbal elements in the caricature, (ii)  list of semantic relations among the elements in the form of triplets (Argument1 -relation- Argument 2), (iii) chains formulated by these relations for message formation (semantic graph path), (iv)  verbal rendition of the chains, (v)   background information (when needed) for better grasping the message of the caricature semantic graph representation of the multimodal message expressed in each caricature, and a full verbalization  of the corresponding message (verbal translation of the multimodal message).

Statistics and annotaction scheme are also included in the release.