Figure 10. Motion saliency maps computed using forward and backward tracks visually uncover motion sinks and sources respectively in cellular videos.

A. Extracted forward superpixel tracks using a target number of 1,000 superpixels overlaid on the first video frame. B. Forward motion saliency map computed using the forward tracks on the left reveals motion highlights the highly dynamic movement of individual cell boundaries which behave akin to motion ‘sinks’. C. Backward motion saliency map computed using the backward tracks produced tracking the video in reverse reveals motion origin or ‘sources’.