Person Re-Identification Datasets for Federated Continual Learning

Person re-identification and federated continual learning has drawn intensive attention in the computer science society in recent decades. As far as we know, this project collects most public datasets that have been tested by person re-identification algorithms, that could be shuffled into different camera visual angles and various time sequences to satisfied the federated learning and continual learning demands across time and space.

• If you use any of them, please refer to the original license.

• If you have any suggestions or you want to include your dataset here, please open an issue or pull request.

• Our Project is a ground-up re-implement of the previous version, awesome-reid-dataset.

Quick Start

Prior federated continual learning applied in Re-ID exists at least 2 issues for datasets:

1. Various camera angles are needed for collaborative training across nodes;
2. Multi-sequences condition, as different tasks, require fulfilled in continual learning.

Before our repository, there not exists an adequate dataset holding various camera visual angles and time sequences to represent the real Re-ID scene. In that case, we provide a solution that mixture and shuffle the current existing public datasets into various angles and sequences to represent reality.

You can quickly split the datasets with default configuration for your experiment as follows before download them:

$ python3 main.py \
    --datasets market1501 prid2011 pku cuhk03 ethz personx \
    --roots ./datasets/Market-1501 \
            ./datasets/prid_2011 \
            ./datasets/pku_reid \
            ./datasets/CUHK-03 \
            ./datasets/ethz \
            ./datasets/PersonX \
    --output ./datasets/preprocessed \
    --split_indice 0.8 0.1 0.7 \
    --task_indice 5 10 \
    --temporal_indice 0.5 3.0 \
    --random_seed 123

Datasets

Dataset	Release time	Identity	Cameras	Sequences	Images	Download
ETHZ	2007	85; 35; 28	1	3	8,580	Google Drive
PRID2011	2011	934	2	1	24,541	Google Drive
Market1501	2015	1501	6	6	32,217	Google Drive
PKU-ReID	2016	114	2	1	1,824	Google Drive
MSMT17	2018	4101	15	1	126,441	Google Drive
DukeMTMC-ReID	2017	1812	8	1	36,441	Retracted
CUHK-03	2014	1467	2	1	13,164	Google Drive
PersonX	2019	1266	12	1	45,792	Google Drive

Contributing

Pull requests are more than welcome! If you have any questions please feel free to contact us.

E-mail: [email protected]; [email protected]

License

Copyright 2021, MSNLAB, NUST SCE

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.