Challenge and Dataset on Large-scale Human-centric Video Analysis in Complex Events (HiEve)

Video Analytics applications like security and surveillance face a critical problem of person re-identification abbreviated as re-ID. Most previous works contributed to the significant progress of person re-identification on still images, but video-based Re-ID is beginning to attract attention. The HiEve dataset, with its diverse camera views and complex events, can provide a good benchmark for video-based Re-ID method. Therefore, we process the HiEve dataset to form the In-scene Video Re-ID Dataset. The resulting dataset is challenging because HiEve contains many discontinuous trajectories.

The clip collections and annotations of our dataset can be downloaded here:

We provide person id annotations for the In-scene Video Re-ID Dataset.

The id annotations is a json file with the following format. The field "id" is the person id number, "clip name" denotes the names of clips containing this group.
person_id{
  "id": int,
  "clip names": [str, str, ...]
}