Fine-Grained Person Re-identification

To study the Fine-grained Person Re-id (FG re-id) problem, we have collected a new FGPR dataset. We collected videos of people wearing uniform clothes. According to the clothes that the people wear, they are divided into three fine-grained groups, namely, “blue”, “white” and “green” groups. As shown in the following Figure, people wear the same clothes in each fine-grained group. The identities in the blue group were captured by camera 1 and camera 2, the white group were captured in camera 2 and camera 3, and the persons in the green group only appeared in camera 4 and 5. The blue group and white group have one common camera view, and the green group has its unique two camera views. Hence, the number of camera views of FGPR dataset is 5. In total, our FGPR dataset includes 134,696 RGB images of 358 identities from 716 sequences.