Enhanced DIIP

Enhanced DIIP (eDIIP) extends the original DIIP dataset with pig-object interaction labels. Label format Each .txt file corresponds to a frame with interaction(s). Box lines are in YOLO axis-aligned format with space-separated values: "class x y w h", where x,y is the box centre, w is the width and h is the height of the bounding box. All values are normalized (0-1) to the image size. - Pig classes (based on the interaction type): 0--biting, 1--snout touch, 2--head touch, 3--body touch. Each pig box encloses the whole animal; the class only indicates the type of interaction the pig is engaged in at that moment (e.g., a box labelled "biting" contains the full pig that is biting an object in that frame). - Object classes: 4--softwood beam, 5--plastic toy, 6--jute bag. - Interactions: Final row lists the interactions as "a,b;c,d" where each interaction is separated by a semicolon. Indices are 0-based and refer to box order in this file; in every pair the first index is the pig, the second is the object. For example 0,3;1,2 -> box 0 (pig) interacts with box 3 (object), and box 1 (pig) with box 2 (object) --- this frame has two interactions. A box may appear in several pairs (e.g. 0,2;1,2 = two pigs interacting with the same object). Usage After unzipping the yolo_labels.zip file in the DIIP dataset, unzip the contents of eDIIP.zip to the same folder. Each label file has the same name as the corresponding image from DIIP (e.g. "pen1_frame_0-02-23.00.txt" contains the interaction labels for "pen1_frame_0-02-23.00.jpg").