Home

Awesome

ClonedPerson

This is the official repository for the ClonedPerson project, which contains the ClonedPerson dataset and code described in our paper "Cloning Outfits from Real-World Images to 3D Characters for Generalizable Person Re-Identification".

<p align="center"><img width=700 src="/img/irregular.png"></p> <p align="center">Fig. 1. Examples of 3D characters in ClonedPerson.</p>

Table of Contents

Pipeline Setup

Please see here.

Dataset Description

The ClonedPerson dataset is generated by MakeHuman and Unity3D. Characters in this dataset use an automatic approach to directly clone the whole outfits from real-world person images to virtual 3D characters, such that any virtual person thus created will appear very similar to its real-world counterpart. The dataset contains 887,766 synthesized person images of 5,621 identities. 3D characters and their counterparts in this dataset are shown in Fig. 1.

Download Links

<!-- Due to the large amount of data, currently only the image used in the experiments of our paper is provided in Google Drive, including 477,555 images (The video is uploading). Images include images and key points position. They can be downloaded from the following links. --> <!-- All data is provided in Baidu Yun Drive, including videos and images (The video is uploading). --> <!-- * [Baidu Yun Drive](https://pan.baidu.com/s/1nfjRzNmxMKddYmVALoXfNw) (code: v6mt) -->

ClonedPerson Dataset

<!-- * [quark](https://pan.quark.cn/s/4b764db4462f) -->

Video_first batch

File Structure

clonedperson
├── clonedperson.tar.gz
│   ├── train
│   │      ├── 700000_s02_c00_f023389.jpg
│   │      ├── 700000_s06_c00_f023413.jpg
│   │      ├── ```
│   ├── test
│   │      ├── gallery
│   │      │      ├── 705228_s06_c02_f005521.jpg
│   │      │      ├── 705624_s12_c02_f002965.jpg
│   │      │      ├── ```
│   │      ├── query
│   │      │      ├── 705227_s02_c02_f001117.jpg
│   │      │      ├── 705624_s12_c00_f003229.jpg
│   │      │      ├── ```
├── videos(uploading...)
│   ├── first batch
│   │      ├── Recordings_scene02
│   │      │      ├── person-im
│   │      │      │      ├── pointsCamera0_1.txt  # This file saved the key points of the camera0_1.mp4
│   │      │      │      ├── deal_camera0_1.txt  # This file saved data extracted from the pointsCamera0_1.txt file every few seconds
│   │      │      │      ├── pointsCamera1_1.txt  
│   │      │      │      ├── deal_camera1_1.txt  
│   │      │      │      ├── ```
│   │      ├── camera0_1.mp4  # This is the video corresponding to file images/scene00/camera0_1.tar.gz(or delete_camera0_1.tar.gz)
│   │      ├── camera1_1.mp4
│   │      ├── Recordings_scene03
│   │      │      ├── person-im
│   │      │      │      ├── pointsCamera0_1.txt  # This file saved the key points of the camera0_1.mp4
│   │      │      │      ├── deal_camera0_1.txt  # This file saved data extracted from the pointsCamera0_1.txt file every few seconds
│   │      │      │      ├── pointsCamera1_1.txt  
│   │      │      │      ├── deal_camera1_1.txt  
│   │      │      │      ├── ```
│   │      ├── ```
│   ├── second batch
│   ├── ```

The filenames are encoded as follows. Take "700000_s02_c00_f023389.jpg" as an example,

<!-- camera*_*_point.txt Data format: image name, the upper left corner of the video x, the upper left corner of the video y, the lower right corner of the video x, the lower right corner of the video y, the distance of head point in the image from the upper left corner x (The following distances are from the upper left corner), the distance y of head point in the image, left shoulder distance x, left shoulder distance y, right shoulder x, right shoulder y, left hand x, left hand y, right hand x, right hand y, left foot x, left foot y, right foot x, right foot y -->

Experimental Results

By training person re-identification models on these synthesized person images, we demonstrate that the model trained on ClonedPerson has a better generalization performance, superior to that trained on other popular real-world and synthetic person re-identification datasets.. The experimental results are shown in the following tables.

<p align="center"><img width=1000 src="/img/experimental.png"> </p>

Contacts

Yanan Wang
Inception Institute of Artificial Intelligence (IIAI)
yanan.wang@inceptioniai.org

Citation

@article{Wang-2022-Clonedperson,
  title={{Cloning Outfits from Real-World Images to 3D Characters for Generalizable Person Re-Identification}},
  author={Yanan Wang, Xuezhi Liang and Shengcai Liao},
   journal={IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2022}
}