person-reid-lib

The pytorch-based lightweight library of person re-identification.

Config

---

Version

python 3.6 or 3.7
pytorch >= 0.4

Install

pip install numpy h5py lmdb
pip install visdom  # Optional. If you don't need a web page visualization, don't install it.

Install pytorch and torchvision

Indicates the folder of the original files and where the unzipped file is placed.

# person-reid-lib/lib/utils/manager.py
self._device_dict = xxxx

Optical Flow

---

Install opencv

pip install opencv-contrib-python    # version 3.4.2.17

Config

# person-reid-lib/lib/dataset/utils.py
DataStoreManager.store_optical_flow = True  # if you want to use optical flow, enable it.

# person-reid-lib/tasks/taskname/solver.py
Solver.use_flow = True

How to run:

---

# image-dataset
cd person-reid-lib_folder
sh script/server_0.sh

# video-dataset

cd person-reid-lib_folder
sh script/task_video.sh

Dataset

---

Image: VIPeR, Market1501, CUHK03, CUHK01, DukeMTMCreID, GRID,

Video : iLIDS-VID, PRID-2011, LPW, MARS, DukeMTMC-VideoReID

Updates

---

2018.12.29 The code of Spatial and Temporal Mutual Promotion for Video-based Person Re-identification is available.

2018.12.26 The initial version is available.

2018.11.19 The code for lib has been released.

Related person ReID projects:

---

deep person reid

MARS-evaluation