Amazon.com, Inc.
Repairing missing frames in recorded video with machine learning

Abstract:

Repairing missing frames in a video includes obtaining video data from an image capture system, applying a first neural network model to the video data to detect that one or more frames are missing, where the first neural network model has been trained to detect missing frames based on training data in which an artificial gap has been introduced. In response to detecting that the one or more frames are missing, a second model is applied to the video data to generate one or more replacement frames. The one or more replacement frames are based on at least a first frame prior to the detected dropped one or more frames, and a second frame after the detected dropped one or more frames. Modified video data is generated using the plurality of frames and the replacement frames.