This document discusses using machine learning for automatic summarization of sport videos. It presents three potential models for classifying success or failure states in video frames: (1) classifying cropped patches of people near the goal post, (2) extracting pose estimation keypoints and classifying those, and (3) combining models 1 and 2 by classifying patches and keypoints together. It then describes in more detail how model 1 works by sampling video frames, detecting people, cropping patches, and classifying the patches. Model 2 and 3 variations are also proposed. The document concludes that GANs have potential for many use cases if developed responsibly.