Image processing using a convolutional neural network to track a plurality of objects

The computing device of 16 comprising a camera and wherein the instructions configure the computing device to:

• present a user interface to receive an appearance choice to apply to the plurality of objects and receive, from the camera, a selfie video image for use as the image and;
• process the selfie video image to produce the updated image using the appearance choice;
• and
• present the updated image to simulate an augmented reality.

A method comprising:

• processing an image comprising a plurality of objects with a convolutional neural network (CNN), the CNN configured to semantically segment the plurality of objects within the image, the CNN comprising a cascaded semantic segmentation model architecture having:
• a first branch providing deep learning of low resolution features;
• and
• a second branch providing shallow learning of high-resolution features;
• wherein the CNN combines respective predictions from the first branch and the second branch to output information comprising foreground/background and object class segmentations.

A method comprising:

• receiving output comprising foreground/background segmentations, object class segmentations and directionality information for each of a plurality of objects semantically segmented by a convolutional neural network (CNN) having processed an image comprising the plurality of objects;
• and
• processing the image to produce an updated image by:
• drawing a gradient of a chosen color over each of the plurality of objects as segmented according to the foreground/background segmentations and object class segmentations, the chosen color drawn perpendicular to a respective direction of each of the objects as indicated by the directional information.