The download contains a zip file with two folders: "data" and "annotations." The data folder contains the video sequences in four sub-folders, "kitchen," "office," "recreation," and "household." Each video sequence is in its own folder, labeled with its activity name, and placed in the appropriate category. For example, a video sequence of somebody spreading peanut butter might be labeled "pb_1" and placed inside the "kitchen" folder. Each folder includes all of the RGB, depth, and thermal image frames.
The annotations folder contains the annotations in two sub-folders, "hands," and "objects." Each of these is organized in the same way as the data folder, and contains the corresonding hand and object-in-interaction annotations, respectively. The annotations are in the form of text files with bounding boxes where each line contains a label, the x1 coordinate (in pixels), the y1 coordinate (in pixels), the x2 coordinate (in pixels), and the y2 coordinate (in pixels) for all labeled items in the image. (x1, y1) represents the bottom left corner of a bounding box and (x2, y2) represents the top right corner of a bounding box.
[Download Dataset] (Coming soon!)