Given (1) an image of a Gymnast from a competition video, (2) the detected 3D coordinates of this player’s skeleton joints at a certain frame (please leave the frame id as a parameter such that we can play with it during the evaluation), and (3) the position of the camera (This is included in "3d->2d convert program" as a parameter in the config file.), you are asked to extract the edges (i.e., contours) of gymnast in the 2D image. Novel and effective approaches are welcome!
In this challenge, we provide images and coordinates for several frames in the same video. The extracted information in the previous frames (e.g., clothes of color and appearance of the background) can be utilized to help the edge detection in the later frames.
The given skeleton coordinates may have about 20cm deviation from the actual position.
You can extract the edges directly from the image without the help of the given skeleton coordinates.
The skeleton coordinates may be useful to generate a bounding box for the player.