User experience platform for connected fitness systems

A method, comprising:

• receiving one or more images that capture an exercise of a user performing an exercise activity;
• providing the one or more images to a machine learning classification network,
• wherein the machine learning classification network includes a 3D-CNN (three-dimensional convolution neural network), a temporal shift module (TSM) network, or a combination thereof, configured to:
• collect feature maps associated with the one or more images across a fixed time window;
• and
• collate and pass the feature maps through a series of convolution layers to output an exercise classification for the exercise performed by the user during the exercise activity;
• and
• receiving, from the machine learning classification network, a prediction of an exercise performed by the user during the exercise activity.

A method, comprising:

• receiving one or more images that capture a pose of a user performing an exercise activity;
• providing the one or more images to a machine learning classification network,
• wherein the machine learning classification network includes a series of encoding layers and decoding layers that generate a pose heatmap having channel-wise peaks that correspond to a pose the user is currently performing within the one or more images;
• and
• receiving, from the machine learning classification network, a prediction of the pose performed by the user during the exercise activity.