Intuitive Physics of Bouncing Spheroids

In a course on computer vision, 3 peers and I created a convolutional neural network to predict the motion of a bouncing spheroid. The idea here was that you could take a video of any ball or ball-like object colliding with a ground plane, preprocess it to some extent, then feed it into this network, which would spit out the location of where the ball will land and what direction it will bounce in after landing (all relative to the camera). We had mixed results, which tracks, since our approach wasn't particularly robust. That said, we got to learn a ton about machine learning, convolutional neural networks, and implementing deep learning networks with PyTorch.

Download the paper here for the full picture.