Object bounding box annotations for the GTEA Gaze+ dataset of the works Learning to recognize daily actions using gaze (Fathi et al., 2012) and Delving into Egocentric Actions (Li et al., 2015). The dataset contains folders for each of the subjects, within each of them folders for actions, and, within each action folder, a folder for each video. The video folder has a name composed of <name of the original video>_<start frame>_<end frame>. Within this folder, a json file for some frames can be found. The json contains two keys: filename and objects. filename refers to the path to the image and objects to a dictionary of objects. The keys of the dictionary are the objects present in the image. Each of the objects values is a list containing bounding box coordinates. Each coordinate list is composed of the ymin, xmin, ymax and ymax values.

{"references": ["Fathi, A., Li, Y., & Rehg, J. M. (2012, October). Learning to recognize daily actions using gaze. In European Conference on Computer Vision (pp. 314-327). Springer, Berlin, Heidelberg.", "Li, Y., Ye, Z., & Rehg, J. M. (2015). Delving into egocentric actions. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 287-295)."]}