AutoTrack: Interactive Visual Object Tracking for Efficient Object Annotations

Efficient video annotation is becoming important with the increasing number of machine learning applications. In this paper, we propose using interactive video object tracking for annotation. Inspired by the VOT short-term tracking challenge, we design an evaluation process to compare the KCF, CSRT, and ATOM short-term trackers. This evaluation process measures the performance and speed of the trackers using these four metrics: accuracy, robustness, initialization time, and processed frames per second. We also modify the open-source CVAT to implement the interactive tracker.