RPIfield: A New Dataset for Temporally Evaluating Person Re-Identification

Abstract

The operational aspects of real-world human re-identification are typically oversimplified in academic research. Specifically, re-id algorithms are evaluated by matching probe images to candidates from a fixed gallery collected at the end of a video, ignoring the arrival time of each candidate. However, in real-world applications like crime prevention, a re-id system would likely operate in real time, and might be in continuous operation for several days. It would be natural to provide the user of such a system with instantaneous ranked lists from the current gallery candidates rather than waiting for a collective list after processing the whole video sequence. Re-id algorithms thus need to be evaluated based on their temporal performance on a dynamic gallery populated by an increasing number of candidates (some of whom may return several times over a long duration). This aspect of the problem is difficult to study with current benchmarking re-id datasets since they lack time-stamp information. In this paper, we introduce a new multi-shot re-id dataset, called RPIfield, which provides explicit time-stamp information for each candidate. The RPIfield dataset is comprised of 12 outdoor camera videos, with 112 known actors walking along pre-specified paths among about 4000 distractors. Each actor in RPIfield has multiple reappearances in one or more camera views, which allows the study of re-id algorithms in a more general context, especially with respect to temporal aspects.