/person-reid

Tensorflow implementation of "An Improved Deep Learning Architecture for Person Re-Identification"

Primary LanguagePython

An Improved Deep Learning Architecture for Person Re-Identification

paper link

Environment

  • Ubuntu 16.04
  • Python 2.7.12
  • TensorFlow 1.2.1
  • CUDA 8.0
  • cudnn 5.1
  • NVIDIA GTX Titan X (Pascal)

Installation

git clone https://github.com/thomaspark-pkj/person-reid.git

Prepare dataset

Download CUHK03 dataset from CUHK Person Re-identification Datasets then extract file.

python cuhk03_dataset.py your_dataset_path

Traininig

python run.py --data_dir=your_dataset_path

Validation

python run.py --mode=val --data_dir=your_dataset_path
Result

Accuracy can be different slightly because image pair is generated randomly in validation dataset.

Accuracy: 0.846667

Test

python run.py --mode=test --image1=your_dataset_path/labeled/val/0000_00.jpg --image2=your_dataset_path/labeled/val/0000_05.jpg
Result
True