License: This dataset is released under CC BY-NC-SA license. Please visit the page for more details.
The training and validation sets for the challenge can be downloaded from following links:
Training dataset: link
Validation dataset: link
Training data has two directories: set1 and set2. Each of which represents a complete independent prostatectomy procedure. Each directory has frames names in the order of their occurrence (e.g. RARP2_frame_1.jpg, RARP2_frame_2.jpg, etc). Hence, participants can use temporal relationships between two consecutive frames. Each frame has its respective label file with the same name (e.g. RARP2_frame_1.txt). Not all the frames in the dataset are labeled for bounding boxes, hence there are some text files in the dataset that will be empty. We advise participants to read files carefully.
Dataset follows the YOLO style of labels. Hence, each row in the text file is a target bounding box and contains five values representing: label, x, y, w, h, exactly in the same order. (x, y) are the coordinates of the center of bounding box while w and h represent width and height of the bounding box, respectively. Values of x, y, w, h are in normalised format. If you want to have an actual coordinate of the bounding box, all x, y, w, h must be multiplied with respective widths and heights of the image.
Obj.names file contains the list of labels, each row in the label list corresponds to each label index in the label files. E.g. if any frame label file has label index 10 then that represents the action in row 11 of the Obj.names file.
The submitted results should be in a single text file with name "submission.txt". The text file should be zipped before uploading it on the submission page. For more details on the structure of the text file, please check the submission page.
Sample submission file format for the result submission is: Link