Labeling every pixel in the video (e.g., distinguishing "road" from "sidewalk").
Most researchers using these specific file IDs are implementing or testing Deep Feature Flow . This framework solves the problem of per-frame processing being too slow for real-time video. 5faafede261a08fdaba46_source.mp4
A heavy network (like ResNet) extracts "deep features" only from select frames. Labeling every pixel in the video (e