Figure 1

The algorithm takes as input an image (INPUT-1) and, optionally, prior shape knowledge (INPUT-2) of the target in certain cases. The user initiates the snake by hovering over the image and clicking on it once. The initial weights of the snake denote the control knots of the cubic spline, functioning as PICS’s hypothesis. Subsequently, we compute losses based on the internal and external energy of the snake model, and optionally, a prior-shape-based loss. The total loss is minimized through gradient descent, leading to updates in the values of spline control knots.