Images of the PerxonX dataset and the original 3D human models used in this work.
Clone or download
Latest commit 4ad87cc Jan 7, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
Scripts Delete RandomPosition.cs.meta Jan 7, 2019
images Add files via upload Jan 7, 2019
README.md Update README.md Jan 7, 2019
example_camera.unitypackage Add files via upload Jan 7, 2019

README.md

Dissecting-Person-Re-identification-from-the-Viewpoint-of-Viewpoint

In this paper, we build a synthetic data engine PersonX that can generate images under controllable cameras and environment. The dataset from PersonX is shown to be as indicative as real-world datasets. Second, based on PersonX, we conduct comprehensive experiment to quantitatively assess the influence of pedestrian viewpoint on person re-ID accuracy. Specifically, the PerxonX engine is built on Unity and its details are as follows.

For researching viewpoint, we use six backgrounds in this version (as shown in above figure), including three pure color backgrounds and three scene backgrounds. There are 1266 hand-crafted identities (547 females and 719 males) and each identety has 36 viewpoints (sampled every from to ), so the total number of images is 273,456 ( ).

fig1


Datasets

We release 3D models of the identeties and the 1266 models can be downloaded from the following links:
Baidu Disk
Google Drive

The package can be imported into any program of Unity directly.

  • "person1-1266.unitypackage" is a Unity package that contains the original 3D models of the identities. You can double-click the package to import is it into your project.
  • The version of Unity used in this work is 2018.1.8f1 Personal that can be downloaded from the Unity Website.
  • The scenes (backgrounds) can be selected or designed based on your demand and there are many free sources in Unity Asset Store. If you want to use the scenes used in this work, please send an email to xxsunzrt@gmail.com.

How to get different viewpoints of the person?

  • Import the scripts in "Scripts" into your project.
  • Create a Game manager(Create Empty) in Unity and create two Camera and one Cube in the manager. fig2
  • Add the scripts on the three components as follows (or directly use the example manager in example_camera.unitypackage) fig3
  • Run the Game Manager in Unity can get the images of a person with different viewpoints.

The original images and unity project will be released ASAP.

If you use this dataset in your research, please kindly cite our work as,

@article{DBLP:journals/corr/abs-1812-02162,
  author    = {Xiaoxiao Sun and Liang Zheng},
  title     = {Dissecting Person Re-identification from the Viewpoint of Viewpoint},
  journal   = {CoRR},
  volume    = {abs/1812.02162},
  year      = {2018}
}