Q: When will you release the dataset?
A: Expected to release first batch data in summer 2018. We are in the process of generating video with various characteristics and annotation.
Q: Who to contact?
A: nuoma at vt dot edu
Q: Any publications to cite?
A: not yet.
Q: What's the point of this dataset?
A: Our goal is to create a software framework that does indoor people detection and tracking in real time. The constraint is indoor environments, works in real time and works in the wild. While MOTChallenge is the golden standard for benchmarking tracking algorithm, the top leaderboards does not necessarialy work well in real world scenarios.
Q: What's future work?
A: This infrastructure provides endless possibilities for various applications. Active learning is a perfect example with the recent advancement in real time pose estimation. Yes, we think a smart building that knows who is doing what is the future, like Sensory As A Service. Or for humanless retail, like Amazon Go, almost entirely computer vison based.